Reading Level Systems Compared: Four Scales, No Conversion
A report lands on your desk. It says the student is reading at Level M, at 650L, at 4.2, and at DRA 28. Four numbers, one child, and a parent in the room who wants to know which one is right.
The honest answer is that all four can be right at once, because the four reading level systems compared here are not four readings of the same instrument. They are four unrelated machines. Two of them are formulas that count things in a text.
Two of them are judgments a person makes while a child reads aloud. None of the four was built to translate into any of the others, and when two of the owners were asked in print whether you may convert between them, both said no.
This article is about the systems themselves: what each one counts, what scale it reports on, and what its own publisher says it cannot tell you. It is not about whether a text is decodable, which is a different question with its own incompatible measures and is covered in Decodable Text Criteria: What Counts, and Who Decides.
The four systems are four different machines
Strip away the branding and there are two families here.
The formulas. Lexile and ATOS are computer programs. You feed in a text, they count features of it, and they return a number. Neither has ever met the student. They are measures of text, and a student only acquires a measure when a separate assessment reports one on the same scale.
The judgments. The Fountas and Pinnell Text Level Gradient and the Developmental Reading Assessment are teaching and assessment tools. A person sits with a student, listens, scores a protocol, and arrives at a level. The text side of both systems is levelled by people reading books, not by a program parsing them.
That difference is the root of nearly every confusion downstream. A Lexile measure is a property of a text that a reader can be matched against. A Fountas and Pinnell level, as the authors use it, is the level at which a teacher should teach a particular student next week.
Those are not the same kind of object, and no arithmetic converts one into the other. A number describing a book and a decision about a child are different species.
Here is what each one actually is.
Lexile: two variables, one number, and seven codes for what the number cannot carry
MetaMetrics runs the Lexile Framework for Reading, and its central design claim is that reader and text sit on one scale. In the company's words, the framework "measures student ability and the complexity of text on the same scale."
What the number is built from
Two, for most texts. Its FAQ on grade levels states that "Lexile measures are based on two well-established predictors of how difficult a text is to comprehend: word frequency and sentence length."
The company's guide to how texts receive measures puts the mechanism plainly: "Generally, longer sentences and words of lower frequency lead to higher Lexile measures. Shorter sentences and words of higher frequency lead to lower Lexile measures."
Two variables, then, and nothing about content, theme, or how much a reader already knows about geckos.
There is one documented exception, and it is worth knowing because it is the same pattern that runs through this whole article. For early reading texts, MetaMetrics adds machinery. Its publisher guide explains that authors of such texts "pay special attention to factors that facilitate comprehension such as easy vocabulary and decoding," and that repetition and patterning matter at that level.
So "to account for these unique factors, the Lexile Text Analyzer evaluates early reading indicators for texts that are 650L and below (for English text only)." There are four, and they are named: Structure (repeated words and phrases), Syntactic (sentence length and word overlap between sentences), Semantic (common versus rare and abstract vocabulary), and Decoding, where "low demand texts have words with fewer syllables and simpler sounds."
Two variables above 650L, six considerations below it. The vendor added the extra machinery exactly where the two variables stop describing what makes a text hard, which is the honest thing to do and also an admission about the two variables.
The scale, and what the prefixes mean
The Lexile Text Analyzer "measures text difficulty on a scale that ranges from beginning texts below 0L to advanced texts above 1600L." On the reader side, a measure "can range from below 0L for early readers to above 2000L for advanced readers," and a reader below zero receives the code BR for Beginning Reader.
That code is a real part of the notation, not a rounding convention. MetaMetrics explains that "a Lexile measure of BR100L indicates that the Lexile measure is 100 units below 0L." A number with a prefix is not a number you can put in a spreadsheet column with the others.
MetaMetrics recommends a working band rather than a point: reading materials "100L below to 50L above" the reader's reported measure, which the company calls the comprehension sweet spot. Note that it is asymmetric, with more room below the measure than above it.
The five things MetaMetrics says its own measure does not do
This is the most useful page in the whole Lexile corpus for a practitioner, and almost nobody reads it. MetaMetrics' guide to how texts receive measures carries a list headed "Lexile text measures do not," and the items are the company's own:
"Indicate text quality, story complexity or style aspects of language."
"Indicate text appropriateness for a student's age or developmental level, background, motivation or interests." MetaMetrics attaches a qualifier to that one, and it belongs with it: "However, book search databases such as the Lexile Find a Book allow students to search for books by Lexile measure and interests."
"Measure text supports such as accompanying pictures found in graphic novels and children's books."
"Comprise the only information that should be used to make reading selections for a student."
"Take into account students' purpose for reading (assignment, pleasure, discovery, research, etc.)."
The grade-level FAQ says the same thing in one sentence: "a book's Lexile measure refers to its text difficulty only. A Lexile measure does not address the content or quality of the book."
Because the number cannot carry any of that, MetaMetrics ships seven two-letter codes that sit in front of it and carry some of it instead. The company's Hub FAQ states the reasoning in one line: "When a book is written for a specific use or style that may be of interest to end users, MetaMetrics provides codes to accompany the measures."
The publisher guide defines them. AD is Adult Directed: "Better when read aloud to a student rather than having the student read independently." HL is High-Low: "Content to engage older students who need materials that are less complex and at a lower reading level."
NC is Non-Conforming: "Good for high-ability readers who still need age-appropriate content." The others are BR, GN for graphic novel, IG for illustrated guide, and NP for non-prose, which covers "poems, plays, songs, recipes and text with non-standard or absent punctuation."
Note that the codes are two-letter prefixes on a number, not a second scale. A book marked HL880L and a book marked 880L carry the same measure and are not interchangeable on a shelf.
A measure that needed no qualification would need no codes. Seven codes is a vendor telling you, in its own notation, where the single number stops working.
What the Text Analyzer is not designed to analyse
Also from the publisher guide, and also rarely quoted. The Lexile Text Analyzer "is not designed to analyze" lists, recipes, poetry, plays and drama, song lyrics, instant messages and text language, student writing, or multiple-choice and fill-in-the-blank questions.
If a student's goal involves a play, a poem, or a worksheet, the instrument that produced the number on the report does not claim to measure it.
On grade levels, MetaMetrics is blunter than the people quoting it
Asked directly whether grade levels can be tied to Lexile measures, the company's answer opens: "There is no direct correspondence of a specific Lexile measure to a specific grade level."
It does publish a table of typical measures by grade, and the caveat attached is explicit. The information "is for descriptive purposes only and should not be interpreted as a prescribed guide about what an appropriate Lexile reader measure or Lexile text measure should be for a given grade."
The same document adds the warning that matters most in a caseload: "Although a student may be an excellent reader, it is incorrect to assume that he/she will comprehend text typically found at (and intended for) a higher grade level." The student "may not have the background knowledge or maturity to understand material written for an older audience."
ATOS: three variables, a grade-level output, and a second field for everything else
Renaissance built ATOS for Accelerated Reader, and unlike Lexile it reports in grade-level units. A book level of 4.4 is a statement in years and months of schooling. The company's own account of what a book level is and is not is unusually frank, and most of this section comes from it.
What ATOS counts
The Common Core text-complexity study describes ATOS precisely, and because that description was published by a third party rather than by the vendor it is worth quoting from there.
ATOS "incorporates two formulas: ATOS for Text (which can be applied to virtually any text sample, including speeches, plays, and articles) and ATOS for Books." Both "take into account three variables: words per sentence, average grade level of words (established via the Graded Vocabulary List), and characters per word."
Renaissance's own description matches, in its own vocabulary: the book level "is determined by the characteristics that most heavily influence readability: average sentence length, average word length and word difficulty level."
The word difficulty component is the part worth knowing about, because it is the only one of the four systems whose owner publishes the size of the corpus behind it. Renaissance states that word difficulty levels "have been determined for over 100,000 words, and are based on analyses of the world's largest database of words from actual books (more than 170,000 books comprising of nearly 2.5 billion words)."
One further design difference, and Renaissance is right to press it: "The ATOS formula used in Accelerated Reader analyses the full text of a book, not merely short passages."
The company's United Kingdom site adds that the formula "is applied consistently to every one of the 26,000+ books we have quizzed," while its United States product page advertises "over 220,000 quizzes." Those are two catalogues counted in two units, and neither number is a count of the other.
Why a Mr Men book and a young-adult novel about racism share a level
Renaissance publishes an explanation of one of its most common support enquiries, and it is the single clearest illustration of what a text-complexity number is.
Teachers kept reporting that their book labels looked wrong. The publisher's answer is that they are not: the level "is determined by the ATOS formula, which is a measure of text complexity and is not affected by the length of a book."
The worked examples are the publisher's own. Mr Greedy, from Roger Hargreaves' Mr Men series, has a book level of 4.4. That is the same level as two of Malorie Blackman's Noughts and Crosses books, and higher than the other two.
Renaissance explains why by quoting the text. Hargreaves' "creative use of slightly unusual words and his habit of stringing them together in long sentences" produces lines like this one, which is quoted in full here because its length is the whole point:
"A huge enormous gigantic colossal plate, and on the plate huge enormous gigantic colossal sausages the size of pillows, and huge enormous gigantic colossal potatoes the size of beach balls, and huge enormous gigantic colossal peas the size of cabbages."
The reverse holds too. Blood Red Road by Moira Young carries a book level of 2.9, because it is "written with very simple language, using stylised grammar and short sentences."
A picture book and a novel about state racism, on the same shelf, at the same number. The formula is not malfunctioning. It is doing exactly what it says it does.
The second field is the admission
Renaissance does not leave it there, and this is the structural point. Every quizzed book also carries an interest level — "a rough guide for teachers, librarians and parents to indicate how appropriate a book is for certain age groups."
On the United Kingdom site, whose band names these are, Mr Greedy is Lower Years and Noughts and Crosses is Upper Years. The publisher adds that it notes "in the summary if a book contains strong language or adult themes."
There is a third field as well. Quiz points are "partly determined by the book's length" — Mr Greedy is worth 0.5 points, Noughts and Crosses 14 — and Renaissance offers that gap itself as "an indication that they won't necessarily be appropriate for the same readers."
So the same vendor publishes three fields because one number could not do the job. Note the parallel with the seven Lexile codes: both formula vendors shipped a second channel to carry maturity and content, having discovered that complexity alone does not.
Renaissance makes its positive case first, and it should be reported alongside the limit: ATOS book levels are "consistent, reliable and objective measures of text complexity," and ATOS "is reliable in what it measures and is an extremely useful guide for librarians, parents and teachers."
Then it states the limit more directly than most publishers would:
"No scientific formula can take into consideration the maturity of the themes addressed or the sophistication of the literary devices employed by the author of a book. Discretion and professional judgment are vital ingredients to successful classroom practice when guiding students to appropriate books."
And, a few lines later: "No formula could possibly identify all the variables involved in matching the right books with the right child."
One thing ATOS has that the others do not
Renaissance describes ATOS as "an open standard," and notes that "the ATOS level of any text can be calculated using a simple online tool," the ATOS Analyser. It is the only one of the four systems in this comparison whose owner uses that phrase about it, and for a practitioner who wants to check a text that has never been quizzed, that distinction has practical value.
Fountas and Pinnell: twenty-six points, ten characteristics, no formula
The F&P Text Level Gradient is the system behind the letter, and it is the least like the other three, because there is no computation in it anywhere.
Fountas and Pinnell describe a gradient "created and refined by Fountas and Pinnell as a teaching and assessment tool over the past thirty years." Each of its "twenty-six points on the F&P Text Level Gradient" runs "from easiest at level A to hardest at level Z," and each represents "a small but significant increase in difficulty over the previous text level."
There is also a Z+ designation on the publisher's website "which refers to the highly complex texts, many of which contain very mature subject matter that students read in high school and college."
The ten characteristics
A level is not a score. It is, in the authors' words, "a composite of ten text characteristics that increase slightly in complexity as you move up the level." The publisher's own FAQ answer, which calls them ten text factors, names all ten:
Genre/Form
Text Structure
Content
Themes and Ideas
Language and Literary Features
Sentence Complexity
Vocabulary
Words
Illustrations
Book and Print Features
Read that list against the two Lexile variables and the three ATOS variables. Sentence Complexity and Vocabulary are the only two entries that a formula is even attempting. Themes and Ideas, Illustrations, and Book and Print Features are precisely the things both formula vendors say their numbers do not capture. They are not an extension of the formulas; they are the complement of them.
What the authors themselves say the gradient is not
The authors make the positive case first, and it is not faint praise: "A gradient of text is a powerful tool for you as a teacher. It helps you in the very challenging task of selecting texts that will challenge your readers and offer them opportunities to learn."
Then comes the passage worth knowing verbatim, which is quoted in the publisher's own teaching materials and almost never in the places levels actually get used:
"A gradient of text is not a precise sequence of texts through which all readers pass. Books are leveled in approximate groups from which teachers choose for instruction. The teacher who recognizes the convenience of the gradient yet reminds herself of its limitations will be able to make good choices and test her decisions against children's behaviors while reading and talking about texts."
Approximate groups. Not a sequence. Tested against behaviour rather than trusted on its own.
The same source is equally direct about the student-facing question, and goes further than most practitioners expect: "It is destructive to measure their own progress by 'moving up levels.' This does not provide the real motivation that consuming and talking about texts would." The post's summary line is as plain as it gets: "To put it simply: a level is a teacher's tool, NOT a child's label."
The authors extend that to the classroom library. Their own resource library states that all children deserve "access to an authentic, rich, and diverse text base - including a robust classroom library where levels have no place," and that the books there "are not there for the purpose of 'practicing reading'" but to create readers "who have the power of choice."
The very next sentence is the one that keeps this from being an argument against levelling, and it is the clearest statement of the authors' actual position: "Levels have a critical place in the hands of teachers who understand their complexity and use them to make good decisions in instruction."
That has a direct consequence for anyone writing goals. The authors of the gradient place the level in the teacher's hands and not in the student's file.
DRA3: twenty-three levels, 102 books, and two readability formulas inside it
The Developmental Reading Assessment, Third Edition is Pearson's benchmark system, and Pearson classifies it as "a criterion-based, authentic benchmark/interim assessment evaluating student reading level and reading behaviors in Kindergarten through Grade 8." The distinction behind the word criterion-based is set out in Norm-Referenced vs Criterion-Referenced.
What the assessment actually produces
Per Pearson's overview, a benchmark administration "helps to determine each student's independent or instructional reading level with an evaluation of three components: reading engagement (optional), oral reading fluency, and comprehension."
The catalogue numbers are specific. Pearson's technical white paper, quoted at length below, states that teachers "can choose from 102 fiction and nonfiction books that are distributed across 23 reading levels." The grades K-3 kit carries Benchmark Assessment Books at Levels A-40; the grades 4-8 kit carries Levels 40-80 "plus Intermediate Bridge Levels 20-38."
So the DRA scale runs from a letter, through the twenties and thirties twice in two different kits, to 80. It is not an interval scale and the numerals are not years, grades, or anything else you can subtract.
The Level Estimator, and the interpolation above level 50
DRA3 added a word-list screener called the Level Estimator, and its development is documented in Pearson's technical white paper. An initial 70-item list was given to 194 students in grades 1 to 5, calibrated with a Rasch item-response model, and reduced to a final 40 items with a split-half reliability of .98.
The Level Estimator total correlated .80 with grade and .94 with a one-minute oral reading passage. Pearson then mapped score bands onto starting levels: a score of 0 to 7 starts at A-3, 20 to 21 starts at 20-24, 38 to 40 starts at 80.
One sentence in that mapping deserves attention. The recommendations were derived from students rated independent on oral reading fluency across benchmark levels 4, 8, 12, 14, 18, 20, 24, 28, 34, 40 and 50 — and then, in the paper's own parenthesis, "scores beyond level 50 were interpolated." The upper third of the DRA scale was not sampled for this mapping. It was extended.
The finding nobody writes: DRA's books were levelled with two readability formulas
Here is the part of that white paper that changes how you read a DRA level.
Fifty-nine of the 102 benchmark books were newly written for DRA3, and the paper sets out the four-phase process. In Phase 2, initial content editing and levelling, the publisher's team reviewed each draft, and the criteria are listed: "Consideration was given to things like the total number of words, total number of sentences, average words per sentence, range of sentence lengths, and Flesch-Kincaid grade level."
Then Phase 3, review by the DRA authors, where the authors checked each book "to ensure consistent leveling by checking the difficulty level of each 100-word section of text using the Fry readability formula."
So the levelling of the texts in a teacher-judgment benchmark system was anchored, at two separate stages, by two readability formulas — one of which, as the next section shows, has a documented identity problem.
This is not a scandal and it is not hidden. It is in a published white paper under the publisher's name. But it does mean that the sharp distinction between formula systems and judgment systems is cleaner in theory than in the materials. A DRA level has a readability formula inside it.
What two trained raters agree about
The same paper reports inter-rater reliability, and the figures belong in any conversation where a single DRA level is being treated as a fact.
Multiple raters scored the same videotaped administrations. Agreement on assigning the performance category for oral reading fluency ranged from 76 to 91 per cent. Agreement on comprehension ranged from 69 to 80 per cent.
At levels 34 and 50, comprehension agreement was 69 per cent. Two trained teachers watching the same child read the same book placed that child in a different comprehension category roughly three times in ten.
Both figures deserve their context, which the paper supplies: "Overall, there was considerable agreement between raters." Internal consistency is stronger still, at coefficient alpha of .84 to .94 for oral reading fluency, median .88, and .78 to .88 for comprehension, median .82, which the paper calls "good to excellent."
Pearson gives its own explanation for why the comprehension side is the looser of the two, and it is the obvious one: correlations are stronger for oral reading fluency "which is expected given the more objective nature of categories rated in Oral Reading Fluency (e.g., rate, accuracy)."
One structural detail is worth knowing alongside that, without reading a cause into it. Each of the 102 books has its own Teacher Observation Guide, but "there are fewer unique Continuums; a total of 14 Continuums are used across all TOGs." The rating scale a teacher applies is shared across a band of levels rather than written per book.
Against an external measure, DRA3 oral reading fluency correlated .72 to .79 with aimswebPlus oral reading fluency across grades 2 to 5, while DRA3 comprehension correlated .62 to .67 with it. The error properties of a fluency score on its own are covered in Oral Reading Fluency Norms, and why a single number deserves a band rather than a point in Confidence Intervals and Standard Error.
There is a common scale, and only two of the four are on it
This is the fact that reframes the whole question, and it is in a document most practitioners have never opened.
When the Common Core standards were released, the developers of the quantitative text-complexity tools volunteered to take part in a comparison study. The result is the Supplemental Information for Appendix A: New Research on Text Complexity, published by the National Governors Association and the Council of Chief State School Officers, the two bodies behind the Common Core.
The work was led by Jessica Nelson of Carnegie Mellon University, Charles Perfetti of the University of Pittsburgh, and David and Meredith Liben of Student Achievement Partners, in association with Susan Pimentel. It is a state-led initiative's document rather than a federal one, and it names no federal agency. The full report sits in the text complexity collection at Achieve the Core.
The six programs that took part
ATOS by Renaissance Learning
Degrees of Reading Power by Questar Assessment, using "a derivation of a Bormuth mean cloze readability formula" on word length, sentence length and word familiarity, reported on a 0 to 100 scale that in practice runs about 25 to 85
Flesch-Kincaid, public domain, using word and sentence length "as proxies for semantic and syntactic complexity"
The Lexile Framework by MetaMetrics
Reading Maturity by Pearson Education, using Latent Semantic Analysis to estimate "how much language experience is required to achieve adult knowledge of the meaning of each word, sentence, and paragraph"
SourceRater by Educational Testing Service, extracting evidence on eight dimensions and accumulating it through three separate regression models for informational, literary and mixed texts
A seventh, the Coh-Metrix Easability Assessor, took part but was excluded from the cross analysis "because it does not generate a single quantitative determination of text complexity" — though the report adds that "it does have use as a tool to help evaluate text systematically."
Notice who is absent. Not Fountas and Pinnell. Not DRA. The common scale is a common scale for computer programs, and the two systems a practitioner is most likely to meet on an elementary report card are not on it.
What the participants agreed to do
"In the final phase of the work, the developers agreed to place their tools on a common scale aligned with the demands of college readiness." Each measure "has realigned its ranges to match the Standards' text complexity grade bands and has adjusted upward its trajectory of reading comprehension development through the grades."
That realignment is why a Lexile band you remember from before 2012 may not match a current one. The bands "have been expanded slightly over the original CCSS scale" at both ends, and the original was published in Lexile units alone.
Here is the resulting alignment, which is the only authorised cross-walk between any of these systems that exists. Each line gives one Common Core grade band and then the matching range in ATOS, Degrees of Reading Power, Flesch-Kincaid, Lexile, Reading Maturity and SourceRater, in that order:
Grades 2-3 — ATOS 2.75 to 5.14; DRP 42 to 54; Flesch-Kincaid 1.98 to 5.34; Lexile 420 to 820; Reading Maturity 3.53 to 6.13; SourceRater 0.05 to 2.48
Grades 4-5 — ATOS 4.97 to 7.03; DRP 52 to 60; Flesch-Kincaid 4.51 to 7.73; Lexile 740 to 1010; Reading Maturity 5.42 to 7.92; SourceRater 0.84 to 5.75
Grades 6-8 — ATOS 7.00 to 9.98; DRP 57 to 67; Flesch-Kincaid 6.51 to 10.34; Lexile 925 to 1185; Reading Maturity 7.04 to 9.57; SourceRater 4.11 to 10.66
Grades 9-10 — ATOS 9.67 to 12.01; DRP 62 to 72; Flesch-Kincaid 8.32 to 12.12; Lexile 1050 to 1335; Reading Maturity 8.41 to 10.81; SourceRater 9.02 to 13.93
Grade 11 to college and career readiness — ATOS 11.20 to 14.10; DRP 67 to 74; Flesch-Kincaid 10.34 to 14.2; Lexile 1185 to 1385; Reading Maturity 9.57 to 12.00; SourceRater 12.30 to 14.50
Read along one line rather than down one measure and the point lands. The grades 2-3 band is ATOS 2.75 to 5.14 and Flesch-Kincaid 1.98 to 5.34.
Both of those are expressed in grade levels, and for the same band of texts they disagree by about three quarters of a school year at the bottom of the band and a fifth of a year at the top. Two tools reporting in the same unit do not report the same number.
The SourceRater column makes the general point even faster. Its range for the grades 2-3 band starts at 0.05 and its range for grade 11 and beyond ends at 14.50, so its numbers look like grade levels and are not grade levels. A number's units are a property of the instrument, not of the digits.
The same text, in five unit systems
The supplemental report includes a worked example, and it is the most useful single paragraph in this whole literature. One passage from Joan Dash's The Longitude Prize, placed in the grades 9-10 band, measured as follows:
ATOS: 10.5
Degrees of Reading Power: 66
Lexile: 1300L
Reading Maturity: 8.67
SourceRater: 10.7
One text. Five numbers. Three of them nominally in grade-level units, and those three span from 8.67 to 10.7, a spread of about two grades on a single passage.
The report's own reading of that example is the opposite of alarmed, and it belongs here in full: "Various readability measures of The Longitude Prize are largely in agreement that the text is appropriate for the grades 9-10 text complexity band."
Which is true, and is the point. At the level of the band, the measures concur. At the level of the number, they do not, and the band is five numbers wide.
The report's own summary of the whole comparison carries the same two halves:
"While there is variance between and among the measures about where they place any single text, they all climb reliably—though differently—up the text complexity ladder to college and career readiness."
And: "No one of the quantitative measures performed significantly differently than the others in predicting student outcomes." A footnote attached to that sentence adds the one distinction the study did find: "Some of the quantitative measures aligned more closely with human judgment regarding where to situate a text within a complexity band, though these measures did not better predict student performance."
So the measures are interchangeable as staircases and not interchangeable as readings. Choosing any one of them gives you a defensible trajectory across a student's career. Comparing two of them on one book gives you two different answers, and neither is the error.
Which is why the report's own recommendation for serious decisions is not to pick the best tool. It is: "In high stakes settings, it is recommended that two or more quantitative measures be used to locate a text within a grade band."
Flesch-Kincaid has no caretaker, and that matters twice over
One footnote in that report deserves to be better known than it is, and for practitioners it matters more than the band table.
When the research team ran their passages through Flesch-Kincaid, they "found no single answer for what the Flesch-Kincaid score was for a specific text. The score depended on which version of the Flesch-Kincaid program was run and how that particular program counted syllables, sentence length, and the like."
The reason is given in the same footnote: "Because Flesch-Kincaid has no 'caretaker' that oversees or maintains the formula, researchers had to make decisions about how to count syllables and sentence length as they programmed the formula."
To get it onto the common scale at all, the team had to adopt one implementation as canonical — "the version of the formula used by Coh-Metrix."
That is already a useful warning about the Flesch-Kincaid readings that appear in word processors and curriculum documentation. But read it next to the previous section and it does something more specific.
Pearson's own technical paper names Flesch-Kincaid grade level as one of the criteria used to level the DRA3 benchmark books, and the paper does not name which implementation. A formula with no caretaker was used at one stage of levelling the texts in a 23-level benchmark system.
None of which makes a DRA level useless. It does mean that the precision implied by the numeral is not in the construction.
The conversion chart question, asked of the owners
Every practitioner has seen a correlation chart with four or five columns, letters down one side and Lexile bands along a row. Here is what happens when you ask the people who own the systems whether that chart is theirs.
Fountas and Pinnell's answer is three words
Their own FAQ takes the question directly — is there a Lexile correlation chart for Fountas and Pinnell levels — and opens the answer with:
"A: No, there is no correlation chart."
The explanation follows: "There are several approaches to determining a text's level of complexity." And the contrast F&P draws is between the two assessments rather than between the two text scales: "A level obtained from a Benchmark Assessment differs from that obtained with a Lexile assessment in that comprehension is a key factor in the Benchmark Assessment."
One detail in that answer is worth checking rather than repeating, and the result is a useful lesson about this whole field. Describing their competitor, Fountas and Pinnell write that MetaMetrics "takes one approach by measuring text complexity with a proprietary computer algorithm that measures sentence length, syllables, and word frequency."
Syllables are not among the two variables MetaMetrics names for its general measure. But they are not an invention either: as the Lexile section above sets out, MetaMetrics' own publisher guide names syllable count under Decoding, one of the four early reading indicators it evaluates for texts at 650L and below.
So the description is loose rather than wrong, and a reader who went only to the headline description of the Lexile measure would have concluded it was wrong.
That is the rule worth carrying. A vendor's account of its own method is primary. A vendor's account of a rival's method is secondary, and should be checked against the rival's own documentation before you either repeat it or correct it.
MetaMetrics published a refutation with arithmetic in it
The other direction has been answered too, and more forcefully. A MetaMetrics FAQ document — it carries MetaMetrics' copyright and trademark notices and directs readers to Lexile.com — addresses the ATOS 2000 scale, a Renaissance conversion of its own grade-based book levels onto a 2000-point scale.
The opening sentence is unambiguous: "Although the ATOS 2000 scale has a range similar to that of the Lexile scale, it is NOT the Lexile scale and SHOULD NOT be interpreted as such."
Then the provenance: Renaissance "developed this scale with NO consultation with MetaMetrics; accordingly, the ATOS 2000 scale should not be used as a way to equate ATOS book levels with Lexile text measures."
And then, unusually for a document of this kind, the arithmetic. MetaMetrics took the ATOS 9.5 level, which Renaissance charts give an ATOS 2000 value of 1258, pulled the books listed at ATOS 9.5 on AR BookFinder, and matched them by title and ISBN to their Lexile measures.
Books in the sample whose Lexile measure was 1260L: one
Average Lexile measure of the sample: 1193L
Range: 1060L to 1340L
Against a recommended working band of 100L below to 50L above a reader's measure, the sample ran as much as 200L below the nominal 1258. MetaMetrics notes that at that level, a 200L gap "represents a 13% difference in forecasted comprehension."
Two system owners have now been asked in print whether their levels convert to a competitor's. Both said no. One of them did the arithmetic to show why.
One publisher does advertise comparisons
Pearson's DRA3 information sheet lists, among the instructional guidance in the kit, "Comparisons to other familiar leveling systems."
That is a vendor-published crosswalk, and it is the exception. The information sheet does not state which systems, on what basis, or with what margin, and this article did not obtain the kit materials to check. If you use that table, it is worth knowing it comes from one of the four parties and has no counterpart from the other three.
So where do the charts actually come from?
Searching for a reading level conversion chart returns a great many of them. On the results reviewed for this article, the charts in circulation came from school districts, regional service agencies, state bodies and educational publishers' supplementary sites — intermediaries assembling a table because practitioners need one, rather than the owners of the scales being assembled.
That was a targeted search and not a census, so treat it as a description of what is easy to find.
That is worth saying plainly, because it is the practical answer to the question you are actually being asked in a meeting. The chart on the wall is somebody's working approximation. It is not a published equivalence, and in at least two of its columns the system owner has stated in writing that no equivalence exists.
None of the four is rated on any federal tools chart
If you wanted an independent verdict rather than a vendor's, the obvious place to look is the National Center on Intensive Intervention, whose charts are the closest thing the field has to a rating authority. What a filled bubble there actually requires is taken apart in NCII Tools Chart Ratings.
Here is the direct answer, from the charts themselves as fetched on 15 November 2026.
Across all three academic charts — academic intervention, academic screening and academic progress monitoring — the following appear zero times as whole-word matches in the full page text: Lexile, ATOS, Accelerated Reader, Fountas, guided reading, DRA, and Developmental Reading Assessment.
One clarification matters here, because the zero is about the four level systems and not about their publishers. Renaissance's own assessments, Star Reading and Star CBM, do appear on the screening and progress monitoring charts, and Pearson's aimswebPlus appears there too. What is absent is any rating of a text-levelling scheme as such.
That is not a criticism of NCII, which rates the tools publishers submit within the domains its technical review committees define. It is the answer to a practical question.
If you are waiting for an independent rating of these four level systems before you rely on one, there is no chart it would currently appear on. The same boundary problem applies to how programs claim evidence, which is covered in Reading Program Evidence Ratings.
Where the quantitative measures stop working, by their own account
The report is candid about three limits, and all three bear directly on caseload decisions.
Kindergarten and grade one. The report states that "texts for kindergarten and grade 1 are still resistant to quantitative analysis, as they often contain difficult-to-assess features designed to aid early readers in acquiring written language." The grade band alignment starts at grades 2-3 for that reason. For the youngest students, the quantitative column is empty by design.
Poetry and drama. "Until such time as quantitative tools for capturing the difficulty of poetry and drama are developed, determining whether a poem or play is appropriately complex for a given grade or grade band will necessarily be a matter of qualitative assessment meshed with reader-task considerations." MetaMetrics says the same about its own analyser, independently.
Narrative fiction in the upper grades. "Research showed more disagreement among the quantitative measures when applied to narrative fiction in higher complexity bands than with informational text or texts in lower grade bands." The report gives the example itself: "some widely used quantitative measures rate the Pulitzer Prize-winning novel Grapes of Wrath as appropriate for grades 2-3," because works like it "often express complex ideas or mature themes in relatively commonplace language."
That is the Mr Greedy problem from the other end of the shelf, and it is the clearest demonstration available that a complexity number is not a measure of what a text demands of a reader.
The report's structural answer is a sequence, not a tool: use a quantitative measure to place a text in a band, then qualitative judgment to place it in a grade within that band. The qualitative dimensions it names are structure, language conventionality and clarity, knowledge demands, and levels of meaning or purpose — which, read next to the ten Fountas and Pinnell characteristics, are substantially the same list.
The Common Core model and the gradient are not rival answers. They are the formula half and the judgment half of one method, and the field's habit of treating them as competing scales is what produces the four-number report.
When several organisations share a vocabulary, check whether they share a scale
This is the pattern this library keeps finding, and reading level systems are a particularly clean instance of it.
Grade equivalents turn out to be a rank dressed as a date, and the publishers who print them say so themselves, in Grade Equivalent Scores. Decodability is claimed in at least four incompatible units with no arbiter, in Decodable Text Criteria.
Pseudoword decoding measures report in units that do not convert, in Pseudoword Decoding Tests. Syllable division rules differ program by program, in Syllable Division Rules. Letter-sound introduction sequences agree on the principle and not the order, in Letter-Sound Introduction Sequences.
Handwriting programs make two decisions before teaching a letter and make them seven ways, in Handwriting Instruction Programs. Mastery criteria publish percentages for constructs the manuals never define, in Mastery Criteria. Progress monitoring decision rules trace to a small number of sources, in Progress Monitoring Decision Rules.
Orthographic processing is a composite nobody has settled, in Orthographic Processing. And the term the whole field now organises itself around has no single owner either, in What the Science of Reading Actually Names.
Reading levels differ from all of those in one respect that is worth noticing. Here, two of the owners have been asked the comparability question directly and have answered it in writing. That is more than most of this field offers, and it is on the public record.
What this changes about reading a report and writing a plan
A sequence, not a decision. The decisions depend on facts about your student that no article holds.
1. Identify which system produced the number before you interpret it. A letter is almost certainly Fountas and Pinnell. A number with an L is Lexile. A decimal in grade-level units is probably ATOS. A bare integer between 4 and 80 is probably DRA. If the report does not say, the only honest next step is to ask, because the four are not readings of one thing.
2. Treat a level from one system as uninformative about another. Not approximate: uninformative, unless you are using a chart whose provenance you can state. Two of the four owners have said in print that no conversion exists.
3. Ask what the number was for. A Lexile text measure describes a book. A Fountas and Pinnell level, as its authors use it, names where to teach next. A DRA level is the output of a protocol with a documented inter-rater agreement range. These answer different questions and only one of them is a property of the text.
4. Put a band around a judgment-based level. Comprehension category agreement between trained DRA raters ran 69 to 80 per cent in the publisher's own study. A level is a reading, not a constant, and a plan that moves a student on a one-level change is reacting to something inside the measurement error.
5. Do not use a level as the goal. The authors of the most widely used level system call that destructive in print, and say the level is a teacher's tool rather than a student label. Goal mechanics that survive this problem are set out in the treatment plan template, and what belongs in the running record in Session Notes and Progress Reports.
6. Name the system and the date in everything you write. "Level M" is not a record. "F&P Level M, benchmark assessment, 14 October" is one, because a future reader can tell what instrument produced it and when. The same discipline that matters for outdated test norms applies here.
7. For a consequential text decision, use two measures. That is the study's own recommendation, not an abundance of caution: two or more quantitative measures to locate a text in a band, then judgment to place it within the band.
8. Tell families the truth, which is more reassuring than it sounds. The four numbers do not disagree because somebody made a mistake. They disagree because they are four different measurements, and the child's reading is the same in all four. That sentence defuses most of the conversations these charts start.
What is out of scope here, and belongs elsewhere
Nothing in this article is legal advice, and nothing here addresses any state or federal obligation about reading assessment or placement. Where a district's own policy requires a particular system or a particular cut, that is a question for the district and, if a dispute is in prospect, for an attorney.
Three further boundaries, so the article is not read for more than it holds:
Whether levelled reading instruction is effective is not assessed here. This article compares what four measurement systems measure and report. No claim is made or implied about outcomes from any of them, and no efficacy verdict is offered on any named product.
Which text a particular student should read is a clinical decision that depends on the student, and this article deliberately gives the sequence rather than the answer.
Decodability is a different property from complexity and has its own measurement problem, treated separately in Decodable Text Criteria. A text can be highly decodable and complex, or neither.
Where the published record runs out
As of 15 November 2026, the following could not be established, and saying so is more useful than papering over it.
No authorised cross-walk exists between the formula systems and the judgment systems. The Common Core common scale covers six computer programs. Fountas and Pinnell and DRA did not participate and are not on it, and there is no comparable study placing them anywhere.
The DRA3 "comparisons to other familiar leveling systems" are listed on Pearson's information sheet, but which systems, on what basis and with what stated margin is not published on the pages checked. The kit materials were not obtained for this article.
Which Flesch-Kincaid implementation Pearson used when levelling the DRA3 books is not named in the white paper. Given that the same formula produced different scores for the same text in the federal study, the implementation is not an idle detail.
How the DRA level mapping behaves above level 50 rests on interpolation by the publisher's own statement, and the underlying sampled levels stop at 50.
The F&P levelling criteria themselves are not freely readable. The ten text characteristics are published, and so is the Factors Related to Text Difficulty chart that defines them — but the full version of that resource requires registration on the publisher's site, so the level-by-level thresholds could not be read for this article.
The two Renaissance catalogue figures are not the same object and should not be added together. The 26,000-plus figure quoted above is the United Kingdom site's count of books it has quizzed; the United States product page advertises "over 220,000 quizzes." Two catalogues, two units, both the vendor's own, and no reconciliation of them is published.
Whether any of the four publishers has revised its position on conversion since the documents cited was not established beyond the pages fetched on 15 November 2026. Absence from a targeted search is not a verdict.
The supplemental report dates from the Common Core implementation period and the band table has not, so far as could be established this run, been superseded by a later multi-measure alignment study.
Frequently asked questions
Can I convert a Fountas and Pinnell level to a Lexile measure?
No. Fountas and Pinnell's own FAQ answers this directly: "No, there is no correlation chart." The two systems measure different things — the F&P gradient is a composite of ten text characteristics including themes, illustrations and print features, while a Lexile measure is computed from word frequency and sentence length.
Conversion charts in circulation are published by districts and intermediaries, not by either system's owner, and should be described that way if you use one.
Which reading level system is the most accurate?
The question does not have an answer in that form, and the Common Core comparison study is explicit about why: "No one of the quantitative measures performed significantly differently than the others in predicting student outcomes." The measures climb reliably but differently.
For the two judgment-based systems there is no comparable head-to-head comparison at all, although Pearson does publish reliability figures for DRA3. What you can say is that none of the four reports on the same scale as the others, so the choice between them is not a choice about accuracy.
Why do a picture book and a young-adult novel have the same ATOS level?
Because ATOS measures text complexity and, in Renaissance's own words, "is not affected by the length of a book." The publisher's worked example is Mr Greedy at book level 4.4, the same level as two Noughts and Crosses books. Renaissance addresses this by publishing a separate interest level field and flagging strong language or adult themes in the book summary. The level and the suitability are two different fields on purpose.
Is a DRA level reliable enough to set a goal against?
Treat it as a reading with a band rather than a constant. In Pearson's own inter-rater study, trained raters watching the same videotaped administrations agreed on the comprehension performance category 69 to 80 per cent of the time, and on oral reading fluency 76 to 91 per cent.
Internal consistency is good, with coefficient alpha medians of .88 for fluency and .82 for comprehension. A one-level change is well inside the range where raters differ.
Does a Lexile measure tell me whether a book is appropriate for a student's age?
No, and MetaMetrics says so in print. Its publisher guide states that Lexile text measures do not "indicate text appropriateness for a student's age or developmental level, background, motivation or interests," and do not indicate "text quality, story complexity or style aspects of language."
The same guide says a measure should not be the only information used to choose a book. The seven Lexile codes exist to carry some of what the number cannot.
Why does the Lexile band for a grade look different from the one I learned?
Because the bands were realigned. As part of the Common Core text-complexity work, every participating measure "realigned its ranges to match the Standards' text complexity grade bands and has adjusted upward its trajectory of reading comprehension development through the grades." The bands were also expanded slightly at both ends of each band. A band remembered from before that work will not match the current table.
Are any of these four systems rated by an independent clearinghouse?
Not on the National Center on Intensive Intervention's academic charts. Checked on 15 November 2026, Lexile, ATOS, Accelerated Reader, Fountas, guided reading, DRA and Developmental Reading Assessment each return zero whole-word matches across the academic intervention, academic screening and academic progress monitoring charts. That reflects which tools publishers submit and which domains the review committees define, rather than a judgment about the systems.
Working at this level of detail across a caseload? The practitioner list gets these comparisons as they publish — assessment instruments, program documentation, and what the published record does and does not say. Join the educator list.




Comments