top of page

Decodable Text Criteria: What Counts, and Who Decides

Writer: Andrea Chernin
Andrea Chernin
Nov 3
20 min read

Updated: 2 days ago

You have a student who needs controlled text, a catalogue full of books labelled decodable, and no way to tell whether the label means the same thing on any two of them. That is not a gap in your knowledge.


There are no settled decodable text criteria, no body that certifies a book as decodable, and no published threshold a text has to clear before a publisher may print the word on the cover.


This is the same shape of problem as the one behind the Orton-Gillingham label, which no organisation adjudicates, and the phrase "the science of reading", which one organisation defines and nobody certifies. The difference is that decodability sounds measurable. It sounds like a percentage. And for twenty-five years a percentage has been quoted at practitioners as though somebody still enforced it.


Here is what the published record actually contains, who published it, and what it leaves to you.


Scope note. Everything below reports what agencies, publishers, rubric-writers and researchers have put in print about textbook content and text selection. None of it describes what any law obliges you, a school or a publisher to do. State procurement rules and the obligations attached to them change, they differ by state, and they belong with an attorney and with your own state agency rather than with an article.

The definition everyone uses contains a word nobody has defined


The working definition in the research literature comes from Stein, Johnson and Gutlohn's 1999 analysis of first-grade basal programs. They describe decodable text as "connected text containing a high percentage of words conforming to the letter-sound correspondences that have previously been introduced."


Read that again and notice what is load-bearing. Not "all". Not a number. High.


Twenty-four years later, in Reading Research Quarterly, Pugh, Hiebert and Kearns put the problem in one sentence: "There is no established definition of the 'high' in 'high percentage.'" That is from a peer-reviewed synthesis published in 2023, and it is still the current state of the field.


The most quoted definition of decodable text in the research literature turns on a word its own authors never quantified.

Two things follow immediately, and both of them matter at your desk. First, decodability is not a property of a book. It is a relation between a book and a sequence of lessons. Second, because it is a relation, the same book is decodable for one of your students and not for another on the same afternoon.


The word is relative in a stronger sense than most people mean


Elfrieda Hiebert made the point as bluntly as it can be made in 2010: "Any text written in English is decodable at some level in that the code never deviates from the alphabetic system."


That is not a quibble. It is the reason the label can be applied honestly to almost anything, and the reason you cannot infer much from its presence.

Where the 80 percent figure came from, and what it actually said


Ask a practitioner where the number comes from and you will usually hear "Texas". That is correct, and the document is still online.


On 26 January 2000, Robert H. Leos, Senior Director of Textbook Administration at the Texas Education Agency, wrote to administrators that "during the recent textbook review, the State Board of Education directed publishers of all conforming first grade reading programs to include text that is on an average 80% decodable based on the cumulative sequence of taught letter-sound correspondences."


Three qualifications in that sentence are almost always dropped when the figure is quoted.


  • It was an average, not a floor. "Each provides reading selections that have an average of 80% decodability."

  • It applied to conforming first grade reading programs submitted in one state's textbook review, not to books in general.

  • It was pegged to "the cumulative sequence of taught letter-sound correspondences" — so the figure is meaningless without the sequence it was computed against.


Pugh, Hiebert and Kearns record the companion figure: California "mandated that texts include 75% decodable words to be considered decodable; Texas mandated 80%." They also record how those numbers were arrived at. States, they write, "have adopted their own standards without evidence."


The part of the Texas letter that nobody quotes is the important part


The letter does not stop at 80 percent. It goes on to describe a second, different measure that the agency reported alongside the first and explicitly did not require.


"Though not required by the State Board of Education, the report includes a column that summarizes the potential for accuracy for each conforming first grade reading program."

And here is how the agency defined that second column: "The potential for accuracy is based on the total percentage of words that students can identify (read accurately) because they have been taught the letter-sound correspondences or have been taught the words as sight words."


So the agency that produced the 80 percent figure published two columns. One counted only words covered by taught letter-sound correspondences. The other counted those plus words taught by sight. It required the first and reported the second as information only.


That distinction has not survived. In the research literature the lesson-to-text-match method is now routinely introduced through the phrase "potential for accuracy" — Saha and colleagues define it that way in their 2020 review, as "the probability that a child will read a given word correctly", operationalised as the match between the teacher edition and the student text.


Hiebert's description of the same model notes, in brackets, that words "can also be taught as sight words and included in the 'decodable' metric."


Two measures that one agency deliberately kept apart have become one measure with two names. And the merged version is the more permissive of the two, because a publisher can raise a sight-word-inclusive decodability figure by declaring more words to be sight words.


If you are comparing two publishers' decodability claims, that is the first question to ask, and almost no catalogue answers it: does your figure count sight words?

There is no standard method, and the people building one say so in print


The clearest statement available on this comes from the team that set out to fix it. In Reading and Writing in 2020, Saha, Cutting, Del Tufo and Bailey open by stating the problem plainly:


"Since no easily accessible, quantitative, word-level metric of decodability exists, we developed a decoding measure."

They expand on it a few paragraphs later: "currently, there is no standard way of determining decoding level of texts, and publishers and researchers use various methods when attempting to create decodable texts."


They group the existing methods into three families — lesson-to-text matching, qualitative levelling, and readability formulas — and conclude that "none satisfactorily capture, in an objective and quantitative manner, a way to pinpoint the decoding difficulty of a word or a passage."


On the percentage specifically, they are just as direct: "the exact percentage of decodable words a text must contain in order to be deemed 'decodable' is not clear."


And there is a second hole, inside the definition itself


This one is easy to miss and it undoes the whole apparatus. The lesson-to-text definition rests on which correspondences have been taught. Saha and colleagues point out that nobody has defined that either:


"There is no research that clearly defines how many exposures or 'lessons' a child requires before a particular GPC becomes decodable."

So the one definition the field agrees on is a match against a set of taught correspondences, and the field has no criterion for when a correspondence counts as taught. You are being sold a ratio whose denominator is undefined and whose numerator depends on a judgement nobody has operationalised.


This is the same structural problem mastery criteria in the program manuals have, and for the same reason: a threshold is published, the construct it applies to is not.

Four incompatible units are in active use


The reason two decodability claims cannot be compared is not sloppiness. It is that the parties making them are measuring different things.


1. Lesson-to-text match, as a percentage


The Texas and California model. Count the words in the text; count those whose letter-sound correspondences have been introduced by that point in a particular teacher's edition; divide. Produces a percentage. Only meaningful relative to one named scope and sequence, and silent on sight words unless the method says otherwise.


2. Phonetic regularity, as a rating


Mesmer's other criterion, as Pugh, Hiebert and Kearns summarise it: "a rating of phonetic regularity, in which GPCs are simple, follow common letter-sound associations, and occur frequently in text." This one is a property of the text alone. It does not reference anybody's lesson sequence, which makes it the only one of the four that would let you compare two books directly — and it produces a rating, not a percentage.


3. A continuous difficulty score, computed per word


In 2020 the Vanderbilt team published a validated decoding measure built from four sub-lexical components: word frequency, the discrepancy between letter count and phoneme count, conditional vowel probabilities drawn from Berndt and colleagues, and consonant blends. It returns a number per word. Their worked examples give airplane a score of 4.071 and jet roughly 1.055.


Across three experiments, the measure explained variance in children's oral reading miscues after controlling for word frequency, and predicted words read correctly per minute after accounting for estimated Lexile scores. The authors describe it as "a promising new measure" and say refinement "is ongoing." It is not a threshold and it was never offered as one.


4. A readability formula standing in for decodability


Saha and colleagues note which formulas the published programs actually use when levelling their materials: Imagine It! applies Flesch-Kincaid, Read Well uses both Spache and Dale-Chall, and Reading Mastery uses Lexile rankings. Three programs, three different formulas, none of them a measure of decodability at all. As the review puts it, their "use for creating decodable text has not been thoroughly vetted."


Four measures, four units: a percentage against one lesson sequence, a regularity rating, a continuous per-word difficulty score, and three borrowed readability formulas. A catalogue that says "decodable" tells you which of these was used in exactly none of these cases.

The only precise definition of decodability in existence is private property


There is a footnote here that is worth knowing about, because it says something about how unadjudicated this word is.


The computational method behind that decoding measure is the subject of United States patent 11,068,659, "System, method and computer program product for determining a decodability index for one or more words", with Laurie Cutting, Ted Hasselbring and Neena Saha named as inventors and announced by Vanderbilt's technology-transfer office.


Its provisional application was filed on 23 May 2017. The patent's own language says the "assigned decodability index indicates an ability of a person to pronounce or sound out the one or more words."


No professional body publishes a decodability threshold. One university owns a patent on a way of computing one.

The rubrics that review curricula do not set a number either


If anybody were going to adjudicate this, it would be the tools districts use to evaluate instructional materials. They do not.


EdReports


The EdReports Foundational Skills review criteria, version 2.1, run to roughly 2,300 words. The word "decodable" appears once as an indicator and twice more in fluency criteria. The words "decodability", "percent" and "percentage" appear zero times.


Indicator 1m reads: "Materials include decodable texts with phonics aligned to the program's scope and sequence and opportunities for students to use decodables for multiple readings."


Look closely at what is being checked. Not alignment to an external standard — alignment to the program's own scope and sequence. The criterion is self-referential by design, which is reasonable rubric-writing and has an awkward consequence: a program that teaches fewer correspondences has less to align to.


A state review tool


Rhode Island's high-quality curriculum materials foundational skills review tool uses "decodable" five times across about 1,200 words and sets no figure. Its criteria are relational in exactly the way the definition is: "Decodable texts are controlled to practice the sound-symbol correspondences that have been explicitly taught", and "Decodable texts included are aligned with the curriculum's content."


Colorado's guidance for selecting decodables


The Colorado Department of Education's page on selecting decodables, dated August 2021, gives four qualitative criteria and no numeric threshold. Its first is the relation again, stated without apology: "Because decodable text is only decodable if children have been taught the particular phonic patterns it contains."


It also asks for something the percentage models cannot express. "This text should not be highly repetitive to the point where students can guess or predict what comes next in a sentence." A text can be 100 percent decodable by lesson-to-text match and still fail that, which is a useful reminder that decodability was never the only criterion anybody cared about.

The federal practice guide uses the word thirty-seven times and never quantifies it


This is the pattern the evidence-rating systems show too, and it is worth doing the arithmetic yourself.


The What Works Clearinghouse practice guide Foundational Skills to Support Reading for Understanding in Kindergarten Through 3rd Grade (NCEE 2016-4008) runs to roughly 57,600 words. Counted as whole words:


  • "decodable" — 37 times

  • "decodability" — zero

  • "80 percent", or any percentage applied to text — zero

  • "controlled" — 102 times, almost all of it in other senses


The guide's definition is purely relational and sits inside the panel's advice for Recommendation 3: decodable text is "text that is written so that students can read it using the letter sounds and high-frequency sight words they have learned."


Note that the federal guide's definition does include high-frequency sight words — the Texas letter's second column, not its first. The two most-cited sources on what decodable means therefore draw the line in different places.


The glossary defines the negative and not the positive


The guide carries a glossary. Under D it defines decoding, digraph and diphthong. It does not define decodable text.


Under N, it does define the opposite: "Non-decodable words are words that the reader is unable to decode."


A federal practice guide that uses "decodable" thirty-seven times, and whose Recommendation 3 carries a Strong evidence rating, defines the word's negation in its glossary and not the word itself.

Recommendation 3 — teach students to decode words, analyze word parts, and write and recognize words — is rated Strong Evidence. Recommendation 4, on reading connected text daily, is rated Moderate Evidence. Neither rating attaches to a decodability percentage, because neither recommendation contains one.

What the publishers themselves claim, and what they leave out


This is where it becomes practical. The claims are not false. They are conditional, and the condition is usually you.


Flyleaf Publishing states the condition outright in its FAQ: "All Flyleaf books are considered decodable if the prerequisite skills on the Scope & Sequence have been taught." That is the honest form of the claim, and it hands the determination to the person doing the teaching.


Benchmark Education's decodable readers page describes texts built from "letter-sounds and spelling and morphological patterns that students have been explicitly taught", plus "a small number of high-frequency words" with irregular spellings.


"Small" is not quantified. The only percentage on the page concerns lesson time rather than text — a Wiley Blevins recommendation that at least half of a decodable reader lesson be spent applying the skill — and the page's three stated quality criteria are instructive, comprehensible and engaging, none of which is a number.


Neither publisher is doing anything improper. Both are describing a relation accurately. What you cannot do is put the two claims side by side and rank them, which is exactly what a practitioner building a text ladder needs to do.


The questions that do discriminate between catalogues are these, and you will usually have to ask them directly:


  • Against which scope and sequence is the figure computed, if there is a figure at all?

  • Does it count words taught as sight words?

  • How many exposures count as "taught"?

  • Is the claim about the whole series, each book, or an average across the series?

  • Is there a list of the irregular words assumed to be known?

What the research shows, with its boundaries attached


This section needs its limits stated in the same breath as its findings, because the evidence here is genuinely thin and the loudest claims on both sides are not supported.


Nobody has shown that a higher decodability percentage produces better reading outcomes. Equally, nobody has shown the reverse. What exists is a small set of studies comparing text types, and they do not converge.


The study everyone cites is more interesting than its headline


Jenkins, Peyton, Sanders and Vadasy (2004), in Scientific Studies of Reading 8(1), 53–85, is the best-known comparison. Shanahan calls it "the best study of the problem" and reports the result plainly: "That degree of decodability made no difference. It made no difference in decoding, word reading, passage reading, or reading comprehension."


The headline figures are usually given as 85 percent against 11 percent. The actual design was staged, and Pugh, Hiebert and Kearns record all six numbers. Across the first, second and last third of instruction, the more-decodable texts were 85, 72 and 80 percent decodable by lesson-to-text match; the less-decodable texts were 11, 40 and 69 percent.


So the contrast narrowed from 74 points to 11 points over the course of the study. The synthesis says so directly: the finding "supports no clear inference — particularly because the study included less decodable texts that were actually in some cases quite decodable."


Both intervention groups outperformed the comparison condition on word recognition, decoding, passage reading fluency and accuracy, spelling and comprehension. The instruction worked. The decodability contrast is the part that did not separate them, and by the end the two conditions were not far apart on the measure in question.


The other studies point in different directions


  • Mesmer (2005) ran a 14-day study contrasting texts at 40 percent and 8 percent lesson-to-text match. Students in the 40 percent condition had greater word reading accuracy and made fewer errors at posttest. A fortnight, a limited set of correspondences, and a much sharper contrast than Jenkins at the end.

  • Price-Mohr and Price (2020), with children aged four to five, found that on reading comprehension tasks, students reading texts with low percentages of decodable words performed significantly better than those reading high-decodability texts; on word reading the difference was marginal, in the same direction.

  • Cheatham and Allor (2012), in Reading and Writing 25, 2223–2246, reviewed the literature and concluded decodability is "a critical characteristic of early reading text", that it makes students more likely to use a decoding strategy, and — importantly — that early reading text should meet multiple criteria with decodability as one of them. They caution explicitly against designing text around decodability alone.

  • Pugh, Hiebert and Kearns (2023), Reading Research Quarterly 58(4), 710–732, pooled effect-size data from three meta-analyses of reading interventions for students with reading difficulty in kindergarten through third grade. The reported effect sizes were .50 for interventions using decodable texts, .49 for non-decodable texts, .41 for no text, and .66 for both kinds together.


Shanahan's reading of the reviews is worth quoting because it is the most defensible summary available: none of them, he writes, "claims that their use improves reading achievement", and "it is most prudent to use both decodables and non-decodables."


What you may not draw from any of this: that decodable text is ineffective, that it is essential, or that any particular percentage is the right one. Three of those four effect sizes sit within a few hundredths of each other, the confidence intervals overlap, and the only condition that stands apart is the one that used both. That is a claim about variety, not about a threshold.


Why the studies cannot settle it anyway


Shanahan names the obstacle, and it is not a design flaw in any one study: "If we all define decodable text in different ways then no one can be very happy with the research."


He adds two confounds that are rarely controlled: word repetition and word frequency. Hiebert's own analysis of older and newer decodables found exactly that divergence — current decodables, she argues, "introduce as many different words with the target phoneme/grapheme(s) as possible", where older series repeated words more. On her counts of unique words per hundred, four older programs ran from 14 to 23.


Two texts can be identically decodable by lesson-to-text match and differ enormously in how many times a student meets each word. One of those properties has a published threshold attached to it. The other, arguably the more consequential for building automaticity, does not.

How the basals actually scored, historically


Worth knowing, because it is the reason the mandates existed and because the numbers are much lower than most people assume.


Saha and colleagues summarise the audits.


In the 1980s, Meyer, Greer and Crummey examined four major basal programs and concluded three of the four had less than 10 percent decodable text by lesson-to-text match.


Stein, Johnson and Gutlohn (1999) analysed several commercially published first-grade basals and found six of seven had little or no relationship to the sequence in their own teacher guides. Across those programs, only 15 percent of the text was decodable on that definition.


Maslin (2007) analysed Harcourt, Houghton Mifflin, McMillan, McGraw-Hill, Open Court and Scott Foresman, and found 16 to 25 percent of the text was not decodable.


Note the direction of Maslin's figure. It is the complement — 75 to 84 percent decodable — which is to say that by 2007 the published basals were landing more or less exactly on the thresholds the two states had set. The audits that motivated the mandates found single-digit and mid-teen percentages. The audit after them found the mandates had been met.


Pugh, Hiebert and Kearns add the sentence that should accompany both facts: "Notably, these empirical data provide no support for the California and Texas mandate levels."

What this changes about how you choose texts


Nothing here argues against using decodable text. It argues against treating the word as information.


Three practical consequences.


Read the scope and sequence, not the label. A decodability claim is a claim about a match. If you do not have the sequence the claim was computed against, you do not have the claim. This is the same move that works on program evidence claims and on alignment claims against the IDA standards: find the document the claim is measured against and read that instead.


Build your own match. The determination the publishers hand back to you is the one you are actually positioned to make, because you know what you have taught. Flyleaf's conditional phrasing is not evasion; it is a description of where the information lives. Keep your own running list of correspondences introduced and irregular words taught, and check a candidate text against your list rather than against the cover.


Expect the ratio to move. Because the denominator is your own instruction, the same book becomes more decodable over a term without anybody reprinting it. That is a feature, and it is the reason a single percentage printed on a series was always a strange thing to want.


And one caution that follows from Cheatham and Allor's conclusion rather than from the critics': decodability was never proposed as the only criterion. Repetition, sentence structure, whether the pictures let a student guess, and whether the text is worth reading all sat alongside it in the original argument. A text ladder built on decodability alone optimises one variable that has no agreed unit.

When several organisations share a vocabulary, check whether they share a definition


This is the ninth or tenth time this site has run into the same structure, and the pattern is now reliable enough to use as a diagnostic.


The Orton-Gillingham label is a claim no organisation adjudicates. "The science of reading" is defined by one body and certified by none. "Dyslexia" carries five different definitions across five bodies.


"Evidence-based" means something different in each of five rating systems. "Aligned" to the IDA standards is checked by nobody outside the publisher. Mastery criteria publish percentages for a construct the manuals never define.


Repeated reading names a procedure each protocol specifies differently. Group size is stated in seven incompatible units across seventeen programs. Grade equivalents and norms mean less than their arithmetic suggests. And the "screener" label covers seven different kinds of object.


Decodability is the version of this problem that hurts most, because it comes with a number. A vague word invites you to ask what it means. A number invites you to assume somebody checked.


The move that works every time: find the document the word is defined in, check whether anybody is obliged to use that definition, and then check whether the people using the word cite it.

What is out of scope here, and belongs elsewhere


Three things, deliberately left out.


What any law requires. Textbook adoption rules, state read-plan policy and the obligations that attach to them are statute's business and they differ by state. The Texas and California figures above are reported as what two agencies published about textbook procurement, at the dates given. Nothing here says what any law requires of you, your practice or a school, and that question belongs with your state agency and with an attorney.


Which books to buy. This article compares what publishers claim, not what their books are worth. That is a different exercise and it needs the scope and sequence in front of you.


How much text practice, and at what cadence. That sits with session frequency and duration and with the published decision rules for progress monitoring, not here.


The same unadjudicated-claim problem runs one level down, into how a long word gets broken up in the first place. Syllable division rules: five programs, five answers compares what each publisher states, and finds that the federal practice guide recommending this instruction names no division pattern at all.


The same unadjudicated-definition problem turns up in what a teacher does the moment a word is misread. Error correction procedures across the published programs compares the three published procedures and reports that the federal foundational skills guide uses the word feedback thirty times and the phrase corrective feedback not once.

Where the published record runs out


Being honest about the gaps is the most useful part of this.


  • No body certifies decodability. Not the International Dyslexia Association, not the Association of Educational Therapists, not a clearinghouse, not a state. The AET Code of Ethics, at roughly 3,000 words, contains "decodable" zero times, "decodability" zero, "phonics" zero and "curriculum" zero. It uses "text" once. The Code governs your conduct, not your materials, and it says so by omission.

  • No current national threshold exists. The two state figures date from 1999 and 2000 and were specifications for textbook submissions in those states' review cycles. Whether any threshold is in force in any state today is a question for that state's agency, and this article does not answer it.

  • The validity of the lesson-to-text method has never been established. Saha and colleagues state this directly, and it is a stronger claim than "the percentage is unclear".

  • Nobody has defined "taught". The quantity of exposure that makes a correspondence count toward a decodability figure is undefined in the literature.

  • The patented index is not a standard and is not positioned as one. It is one validated measure whose refinement its authors describe as ongoing.

  • Publishers' internal methods are mostly not published. Saha and colleagues make the same observation about qualitative levelling systems — that the reasons a text receives a given level "is not available (even from the publishers)" for the two most prominent systems.


If you need a defensible decodability figure for a specific text, there is currently no public instrument that will give you one. What you can do is state your own criterion, apply it consistently, and write it down — which is more than the field has managed.

Frequently asked questions


Is there an official percentage a book must hit to be called decodable?


No. There is no current national or professional threshold. Texas's State Board of Education directed publishers of conforming first-grade programs to average 80 percent decodability in a textbook review announced in January 2000, and Pugh, Hiebert and Kearns record California at 75 percent, describing both as adopted "without evidence". Both were procurement specifications in one review cycle each, and the same synthesis notes the research supports neither level.


Who certifies a book as decodable?


Nobody. No professional body, clearinghouse or accreditor certifies decodability. The two most widely used curriculum review rubrics both check decodable texts only for alignment to the program's own scope and sequence, and neither states a percentage. The determination is left to the publisher's own method, which is usually unpublished, and ultimately to the person teaching.


Can I compare two publishers' decodability claims directly?


Generally not, and this is the practical heart of the problem. At least four incompatible units are in use: a percentage computed against one named lesson sequence, a rating of phonetic regularity, a continuous per-word difficulty score, and borrowed readability formulas such as Lexile, Spache or Flesch-Kincaid. Unless both publishers name their method, their sequence and whether sight words count, the two figures are not on the same scale.


Does a higher decodability percentage mean better results for my student?


The published evidence does not support that claim, and it does not support the opposite either. The best-known comparison found no difference in decoding, word reading, passage reading or comprehension between conditions, while a pooled analysis of three meta-analyses reported effect sizes of .50 for decodable texts, .49 for non-decodable and .66 for both together. No individual prediction follows from group comparisons, and no study licenses a target percentage.


Why do the same books count as decodable for one student and not another?


Because decodability is a relation, not a property. The standard definition measures a text against the letter-sound correspondences that have already been introduced, so the figure changes with the reader's instructional history. A federal practice guide puts it as text students can read "using the letter sounds and high-frequency sight words they have learned" — which is why the ratio for a given book rises over a term without the book changing.


What should I actually ask a publisher?


Five questions: against which scope and sequence is the figure computed; does it count words taught as sight words; how many exposures count as "taught"; is the claim per book, per series or an average; and is there a published list of the irregular words assumed known. A publisher who answers all five has given you something usable. A cover saying "fully decodable" has not.

Working in this field?


Illuminate Educational Therapy Group publishes this resource library for educational therapists, learning specialists and executive-function coaches — practitioners doing this work, not parents researching it. If the comparisons here are useful, join the educator list and new analyses arrive as they publish.


Recent Posts

See All
bottom of page