Oral Reading Fluency Norms: What the Table Does Not Say
Updated: 4 days ago
You have a words-correct-per-minute score in front of you, a grade-level table open in another tab, and a parent who wants to know whether the number is a problem. The table will give you an answer. It will also, quietly, answer a different question than the one you asked.
Oral reading fluency norms are the most widely reproduced numbers in reading assessment and among the least widely read. The table practitioners use comes from a 2017 technical report out of the University of Oregon.
The decision rule practitioners apply to it does not come from that report at all. The two have been circulating together for so long that almost nobody notices they are separate documents with separate claims and separate silences.
This is not a complaint about the norms. They are careful work, published for a specific and defensible reason, and the authors describe their method in enough detail that you can audit it yourself. It is a piece about what the table carries and what it leaves to you — because once you have read the method, you will read the number differently, and you will say something different to the parent.
What the 2017 compiled norms actually are
The document is Technical Report #1702, An Update to Compiled ORF Norms, by Jan Hasbrouck and Gerald Tindal, published by Behavioral Research and Teaching at the University of Oregon in 2017. It is seventeen pages. It is free, and the National Center on Intensive Intervention hosts a copy alongside its own tools charts.
It is also the third in a series: Hasbrouck and Tindal published compiled norms in 1992 for grades 2 through 5, drawn from roughly 45,000 students, and again in 2006 for mid-grade-one through the end of grade eight, drawn from roughly 250,000.
The word "compiled" is doing real work in that title, and it is the first thing to understand.
The 2017 table is an average of three published tables
The 2017 norms were not produced by testing a sample of students. They were produced by averaging the percentile tables that three commercial assessments had already published. The report states the method plainly:
These new ORF data files were compiled from technical documents that reported norms specific to each individual assessment. These three sets of assessment-specific norms, rather than raw scores from those three assessments, were then averaged to compile this new set of compiled ORF norms.
The three sources, with the school years their data came from, are DIBELS 6th Edition (2009–2010), DIBELS Next (2010–2011), both from the University of Oregon DIBELS Data System, and easyCBM (2013–2014). The report's Table 3 gives the number of scores behind each source table, and the total comes to 6,886,582.
That total is the sum of the samples behind the three source tables. It is not a sample the authors drew, and the averaging was done on the percentile values rather than on the scores. Table 3 also shows how unevenly those samples sit next to each other: in the fall of grade two, DIBELS 6th Edition contributes 637,017 scores, DIBELS Next contributes 4,231, and easyCBM contributes 500.
Those three sources are also not three independent instruments, which is a point the report does not make and which matters a great deal. More on that below.
Grades one through six, and one empty row
The 2017 table covers grades 1 to 6 at the 10th, 25th, 50th, 75th and 90th percentiles, for fall, winter and spring. The grade one fall row is blank, because none of the three sources administers an oral reading passage to first graders in the autumn — which is sensible, and which means there is no compiled fall benchmark for a first grader at all.
What the report says the norms are for
This is the sentence that should change how you use the table, and it sits in the summary rather than the introduction:
These compiled norms were published to prevent a low-performing school or district from setting benchmark goals for their students at a level that was lower than it should be.
The compiled norms exist as a floor against locally-set expectations drifting downward. The report explains the problem they were built to solve: if a district norms its own expectations on its own low-performing population, "students at-risk for academic failure may be identified as low risk when their performance is compared to norms of other low performing students."
That is a systems-level purpose. It is a good one. It is not the same thing as a target for one child on your caseload, and the report never offers the table as one.
What the report does not claim
Three words you would expect to find in a norms document do not appear in Technical Report 1702 at all: "representative," "nationally," and "caution." The word "national" appears twice — once in the University of Oregon's non-discrimination boilerplate, and once in a reference to a National Center for Education Statistics data file.
The authors do not describe their compilation as nationally representative. Reproductions sometimes do it for them. One widely circulated copy, hosted by a literacy training institute, is filed under a name containing the words "NATIONAL NORMS" even though the document inside it never uses the phrase.
The report's own summary makes a narrower and more honest claim: that the three studies across twenty-five years "can give educators a perspective on the stability of ORF scores across materials and grades."
Stability is the claim. Representativeness is not.
The newest data in the 2017 norms was collected in 2013–14
Work the dates forward. The most recent of the three source data sets is easyCBM's, from the 2013–2014 school year. The two DIBELS data sets are from 2009–10 and 2010–11. As of this writing, the freshest number in the "2017 update" describes reading performance from more than a decade ago, and the oldest describes performance from before the Common Core State Standards were implemented in most states.
The report does not hide this. It prints the years. The problem is that the table does not print them, and the table is what gets reproduced.
If you have read our piece on what actually makes a set of test norms too old, you will recognise the shape of this. The ten-year rule practitioners cite for norm recency appears in none of the documents they cite it from, and the right question is never simply how old the norms are.
It is whether anything has changed in the population or the construct that would make the old distribution the wrong comparison. For oral reading fluency across a decade that included a pandemic, that is a live question and not a rhetorical one — and neither this report nor any compiled successor answers it, because there has been no compiled successor.
Three sources, three administration conventions, one averaged table
Here is the finding that most changes what a WCPM number means, and it requires reading a second document from the same lab.
How many passages make a score? The publishers disagree. In Best Practices in Oral Reading Fluency Administration, presented at the 2013 National Association of School Psychologists convention by Daniel Anderson, Julie Alonzo and Gerald Tindal — the same Tindal, from the same research group that published the norms — the divergence is stated directly:
For example, the developers of easyCBM suggest basing decisions on a single ORF probe (Alonzo & Tindal, 2012), while the DIBELSNext developers suggest administering three probes in succession, taking the median score (DibelsNext, 2011).
So two of the three source tables averaged into the 2017 compiled norms rest on different administration conventions. One is built from single-probe scores. The other is built from medians of three.
Technical Report 1702 does not mention administration at all
The words "median," "probe," and "administration" appear in the 2017 report zero times. There is no instruction on how many passages to give, no statement of which convention each source used, and no adjustment for the difference. The report gives you the percentile table and the method by which the three tables were averaged, and it stops there.
This matters because a median of three and a single probe are not interchangeable measurements of the same thing, and the authors of the administration paper say why: a median-score approach exists partly "to help alleviate test form effects" and partly "to produce more stable estimates of students' fluency." A single probe carries whatever the passage happened to be.
And the chart's own guidance asks for a third convention
The most widely used interpretive rule for these norms appears on the Reading Rockets fluency norms chart page, which hosts the 2017 table:
Students scoring 10 or more words below the 50th percentile using the average score of two unpracticed readings from grade-level materials need a fluency-building program.
Read the conditions in that sentence, because they are easy to skim past. It asks for the average score of two unpracticed readings from grade-level materials. Not one probe. Not the median of three. The average of two.
That is a third convention, and it matches neither of the two that produced the data in the table it sits above. It also does not appear in Technical Report 1702 — the phrases "10 or more," "fluency-building," and "grade-level materials" are absent from the report entirely.
The rule is credited to the authors' 2006 article in The Reading Teacher, which is a different publication from the technical report that carries the 2017 numbers.
So when a colleague says "the Hasbrouck and Tindal norms say he needs a fluency program," there are two documents in that sentence and they are not the same one. The table came from the 2017 technical report. The threshold came from a 2006 journal article, by way of a chart page, and it carries an administration requirement most practitioners do not meet.
The three sources are two lineages, and the word "DIBELS" has changed hands
If a school report lands on your desk saying a student's DIBELS oral reading fluency score was 84, you need to know which DIBELS. There are now two instruments with a claim on the name, they are published by different organisations, and the history is stated in a footnote that almost nobody reads.
Here it is, from Acadience Reading's own benchmark documentation:
Acadience Reading K–6 is the new name for the DIBELS Next assessment. ... The DIBELS Next copyrighted content is owned by Acadience Learning Inc. The DIBELS and DIBELS Next registered trademark was sold by Acadience Learning Inc. to the University of Oregon (UO) and is now owned by the UO.
Read that carefully, because the consequences are not obvious.
What follows from the name sale
The instrument formerly called DIBELS Next is now called Acadience Reading, and its content is owned by Acadience Learning. The name DIBELS belongs to the University of Oregon, which publishes DIBELS 8th Edition — a different instrument, with its own passages, its own norming and its own benchmark goals. The National Center on Intensive Intervention's tools charts list the instrument as "Acadience Reading (aka DIBELS Next)" for exactly this reason.
So two of the three "widely-used commercially available ORF assessments" averaged into the 2017 compiled norms — DIBELS 6th Edition and DIBELS Next — are successive editions of one instrument from one publisher, Dynamic Measurement Group. Only easyCBM, from the University of Oregon, is a separate lineage. The compilation averages three tables from two instrument families, weighting each table equally.
And when, later in this piece, we set the compiled 50th percentile beside an Acadience benchmark goal, remember that Acadience is DIBELS Next. One third of the compiled norm and the Acadience benchmark come from the same passages. They differ because one is a percentile and the other is a predictive cut score, not because they are measuring different children on different material.
Why the sixth grade numbers look wrong and are not
Dynamic Measurement Group's own DIBELS Next benchmark goals document anticipates the obvious objection to its own table, and the answer is the most useful sentence in it:
Why are the sixth grade benchmark goals lower than the fifth grade goals? Answer: The difficulty level of the passages used for DORF and Daze changes by grade, so composite scores and benchmark goals can't be directly compared across grades. The difficulty level of the passages increases by grade in a roughly linear fashion. However, student performance increases in a curve, with the most growth occurring in the earlier grades, and slower growth in the upper grades. Between fifth and sixth grade, the difficulty level of the materials increases at a faster rate than student performance, so benchmark goals are lower in sixth grade than in fifth.
Words correct per minute, then, is not one unit running down the page. It is a rate on this grade's passages, and the passages get harder. The publisher says the scores "can't be directly compared across grades."
The compiled table lays grades one through six in a single column headed WCPM, which invites precisely the comparison the publisher rules out.
It also produces the artefact noted earlier: the compiled 50th percentile for spring is 146 in grade five and 146 in grade six, and four grade six cells moved down between 2006 and 2017. Neither is a flaw in the compilation. Both are what happens when you stack rate scores from progressively harder passages into one column.
The practical rule is narrow and firm. A fluency score compares a student to other students in the same grade on comparable material. It does not tell you how many words of progress a child has made since last year unless the passages are of equivalent difficulty, and across grade boundaries they are not designed to be.
A norm, a benchmark goal and a cut point for risk are three different objects
If the compiled norms are descriptive, what is prescriptive? Two other systems publish numbers that look exactly like the compiled percentiles and are built to answer a completely different question.
What Acadience publishes
Acadience Reading's benchmark documentation — the same document that carries the name history above — is explicit about its own category:
Acadience Reading benchmarks are empirically derived, criterion-referenced target scores that represent adequate reading skill for a particular grade and time of year.
Criterion-referenced, not normative. And the criterion is named: "A benchmark indicates a level of skill at which students are likely to score above the 40th percentile on any high-quality reading assessment and achieve the next Acadience Reading benchmark or reading outcome." The document attaches a probability to that: for a student at or above benchmark, "the overall likelihood of achieving subsequent reading benchmarks is approximately 80% to 90%."
It also attaches a caveat in the same breath, and the caveat is the kind of sentence the compiled norms never need because they make no prediction: "A score at or above the benchmark indicates that the odds are in the student's favor of achieving the next benchmark, but it is not a guarantee."
The underlying research is dated in the document too. The DIBELS Next benchmark goals, cut points for risk and composite score "were developed based upon data collected in a study conducted during the 2009–2010 school year," with the Group Reading Assessment and Diagnostic Evaluation as one external criterion for a subset of 1,306 students across grades K to 6.
Those are the same school years feeding two of the three compiled-norm source tables, which is not a coincidence: it is the same data era.
What DIBELS 8th Edition publishes
DIBELS 8th Edition publishes two cut scores per measure rather than one, and the research brief explains both:
The first score, called the risk cut-score can be used to classify students who are at risk for reading difficulties, including dyslexia. The second score, called the benchmark goal, can be used to classify students who are at some risk for not meeting proficiency goals versus those who are on track for meeting proficiency goals.
Both were derived by receiver operating characteristic analysis against an external criterion — DIBELS Next composite percentile ranks in kindergarten, and Iowa Assessment Total Reading percentile ranks in every grade above it.
The brief reports the resulting classification accuracy measure by measure and grade by grade. For oral reading fluency in grade two, for instance, it reports areas under the curve of .89 to .93 for the risk cut-score and .86 to .91 for the benchmark goal.
Why the compiled norms report no sensitivity at all
A percentile makes no prediction. It says where a score falls in a distribution, so there is no later outcome it could be sensitive or specific to. That is not a defect in the compiled norms — it is a property of what a norm is.
But it means a benchmark goal and a 50th percentile are not two estimates of the same quantity, and treating them as interchangeable is the single most common error in how these tables get used.
This is the same distinction that runs through our comparison of progress monitoring tools: a benchmark is not a percentile, and switching systems mid-caseload breaks the graph. It is also the distinction that makes grade equivalent scores so persistently misread, for a structurally identical reason — a descriptive statistic gets handled as though it were a target.
The same grade, the same season, three numbers
Set the three systems beside each other for the end of the school year and the gap is not subtle. These are the published figures, read from each source document:
Grade 2, spring. Compiled 50th percentile: 100 WCPM. DIBELS 8 minimum for core support: 94. Acadience benchmark: 87.
Grade 3, spring. Compiled 50th percentile: 112. DIBELS 8 minimum for core support: 114. Acadience benchmark: 100.
Grade 4, spring. Compiled 50th percentile: 133. DIBELS 8 minimum for core support: 125. Acadience benchmark: 115.
Grade 5, spring. Compiled 50th percentile: 146. DIBELS 8 minimum for core support: 137. Acadience benchmark: 130.
Grade 6, spring. Compiled 50th percentile: 146. DIBELS 8 minimum for core support: 141. Acadience benchmark: 120.
Twenty-six words correct per minute separate the highest and lowest answer for a sixth grader finishing the year. None of the three numbers is wrong. They are answers to three different questions: where is this score in a distribution of scores from 2009 to 2014, what score predicts being on track against the Iowa Assessment, and what score carries roughly eighty to ninety percent odds of meeting the next Acadience benchmark.
Two caveats on reading that list across rows rather than down them. The Acadience column and one third of the compiled column come from the same instrument under two names, for the reason given above. And the compiled 50th percentile is 146 in both grade five and grade six, which the DIBELS Next documentation explains as a passage-difficulty effect rather than a plateau in children's reading.
The score on your protocol has a standard error, and nobody prints it beside the percentile
This is where the published record is most generous and the reproduced table is most silent.
What the same lab reports about single-probe error
The administration paper from Behavioral Research and Teaching reports predicted absolute standard errors for a single test form on a single occasion, by grade. Reading from its Table 3: grade 1, 10.32 words correct per minute; grade 2, 13.30; grade 3, 16.23; grade 4, 13.85; grade 5, 12.80.
The authors then do the arithmetic for you:
So, for a student at grade 1 who reads 75 words correct, we can be 95% confident that his or her true score lies in the range 75 ± (10.32 × 1.96) ≈ 55 to 95. Using this same formula, we can see how our range increases at grade 3, where the standard error is 16.23. At grade 3, we would be 95% confident that a students true score would lie roughly in the range of 43 to 107.
A sixty-four-point interval, at grade three, from one passage.
What the wider literature reports
The administration paper summarises two other estimates. Poncy, Skinner and Axtell estimated the standard error of measurement of a single passage at 12 or 18 words depending on how the passage was constructed, falling to 5 or 7 words when three passages were used.
And Christ and Silberglitt's 2007 study in School Psychology Review, using a large multi-year sample across grades 1 to 5 with a median-score administration, found a median standard error of measurement of 10 words per minute, with estimates ranging from 4 to 15.
Adding a second form, the Oregon group found, "reduced the estimated absolute standard error by approximately 3 words per minute."
Put the error next to the percentile gaps
Now compare those intervals to the resolution of the table. In the spring of grade three, the compiled percentiles run 63, 91, 112, 139, 166 for the 10th, 25th, 50th, 75th and 90th. The gaps between adjacent published percentiles are 28, 21, 27 and 27 words correct per minute.
A single-probe ninety-five percent interval at grade three spans roughly sixty-four points. It is wider than any two adjacent percentile bands in that row put together.
A single passage, in other words, cannot reliably place a third grader in a percentile band — not because the table is bad, but because one minute of reading is a small sample of behaviour and the table prints to a resolution the measurement does not support.
And compare it to expected growth
The administration paper also reports expected weekly growth at the 50th percentile on easyCBM passage reading fluency: 0.87 words per minute per week in first grade, 0.63 by fifth grade.
Divide. A median standard error of 10 words per minute is worth about eleven weeks of expected growth at grade one, and about sixteen weeks at grade five. Roughly a term either way, inside the error of a single score. That is the number to hold in mind the next time a graph appears to jump.
Technical Report 1702 cites the standard-error paper and reports none of its own
The word "error" appears twice in the 2017 norms report. Once, describing the scoring convention — the errors an examiner marks while the student reads. And once in the reference list, in the entry for Christ and Silberglitt's paper, whose entire subject is the standard error of measurement of oral reading fluency.
The report cites that paper to support a statement about expected rates of growth. It does not report a standard error, a reliability coefficient or a confidence band for any figure in its own tables. The word "confidence" does not appear in the document.
Compare that with the primary source behind one of its three inputs. The easyCBM 2014 norms technical report by Saven, Tindal, Irvin, Farley and Alonzo volunteers the point unprompted:
Such slight variations between the group norms and the individual percentile ranks should not cause alarm, but instead remind teachers and administrators that all measurement contains some minor amount of error: score values are best represented in confidence bands (the range of scores within which we can be confident a specific score exists), rather than as static numbers.
One of the documents averaged into the compiled table tells you to use confidence bands. The compiled table has no column for one.
How easyCBM's contribution was built
While we are in that report, it is worth knowing what the "500" in the compiled norms' Table 3 refers to. easyCBM's 2014 norms were constructed by proportionately and randomly sampling 500 students from each of four regions — Midwest, West, Northeast and Southeast — with the published summary figure being the median of the four regional values.
Where a region had too few students for a measure and season, the report says, "we weighted the population in order to reach 500 (so long as the weight did not exceed two)."
Follow the chain for a single cell of the compiled table. A percentile from easyCBM is the median of four regional percentiles, each from a 500-student sample that may have been weighted. That percentile is then averaged, unweighted, with a percentile from a DIBELS table built on hundreds of thousands of scores. Three aggregations stand between a child reading aloud and the number on the chart.
One naming mismatch, for completeness
easyCBM does not call its measure oral reading fluency. Its own norms report calls it Passage Reading Fluency throughout and never uses the phrase "oral reading" at all. Technical Report 1702 refers to it as "the easyCBM ORF assessment."
The measures are the same kind of task — a timed oral reading of an unpracticed passage — and nothing here suggests otherwise. But if you go looking for easyCBM ORF norms you will not find them under that name, and knowing why saves a frustrating ten minutes.
Added 9 October 2026. Because this question arises on every instrument and not only on a fluency probe, it now has a piece of its own: what the confidence band on a score report actually is, why its width is a setting rather than a property of the student, and why comparing two scores needs an error term no report prints.
What happens when the table is reproduced
The 2017 report is three pages of tables inside a seventeen-page document. What circulates is the tables. That makes reproduction fidelity a practical question, and it does not survive inspection.
A state agency's copy of the 2006 table
The Colorado Department of Education publishes a guide for assessing fluency rates and accuracy built around the 2006 Hasbrouck and Tindal table. It is a clean, useful one-page sheet, and it includes something the 2017 table does not have at all: an average weekly improvement column, in words per week, for every grade and percentile.
It also disagrees with the authors' own reproduction of the same 2006 table. In the Colorado sheet, every grade eight winter figure is identical to the grade eight spring figure at all five percentiles — 199 and 199, 177 and 177, 151 and 151, 124 and 124, 97 and 97.
In Table 2 of Technical Report 1702, where the authors reprint their own 2006 norms, those winter values are 199, 173, 146, 115 and 84. The grade seven winter 90th percentile reads 195 in the Colorado sheet and 192 in the authors' report — and 195 happens to be the grade six winter 90th percentile in both.
The pattern is consistent with a column-copying slip in the state document rather than with two genuine versions of the 2006 norms. Either way, the rule that follows is simple: for the 2006 figures, use the authors' reprint, not a secondary sheet.
The authors' own report has a typographical error too
In fairness, so does the primary document. In Table 2 of Technical Report 1702, the grade three spring 25th percentile for the 2006 norms is printed as 787. In Table 5 of the same report, which compares 2006 with 2017 cell by cell, the same figure reads 78.
The Colorado sheet also reads 78. The intended value is clearly 78, and a practitioner copying from Table 2 without cross-checking Table 5 would carry away a number that cannot exist.
This is the reason to read a technical report rather than a chart, and then to read it twice. The useful habit is one we have argued for before in the context of achievement test qualification levels: go to the issuing document, and when two parts of the same document disagree, say so rather than picking one silently.
Grades seven and eight were in the 2006 table and are not in the 2017 one
The 2006 compiled norms ran from the middle of grade one to the end of grade eight. The 2017 compiled norms stop at grade six. The report says so and does not explain the change beyond noting that "these current scores only provide norms through grade 6."
For practitioners whose caseloads include middle schoolers, that is not a footnote. It means the most recent compiled oral reading fluency norms do not reach your student, and your options are these:
The 2006 compiled table, which does cover grades 7 and 8, and which rests on data roughly two decades old.
A publisher's criterion-referenced benchmarks. DIBELS 8th Edition publishes oral reading fluency benchmark goals for every grade from one through eight, with classification accuracy reported by grade in the research brief. Its grade eight end-of-year minimum for core support is 135 words correct, with a blue "negligible risk" threshold at 159.
Local comparison, which is what the original curriculum-based measurement researchers recommended in the first place, and which the 2017 report describes at length before explaining its hazards.
The growth column is the other casualty. Average weekly improvement figures exist in the 2006 publication and not in the 2017 one, so a practitioner setting a rate-of-improvement goal from compiled data is working from the older table whether they realise it or not.
Our piece on intervention session frequency and duration makes the adjacent point about published dosage: the specification in the manual answers a fidelity question, not a question about what your weekly caseload can deliver. Growth norms behave the same way.
The strongest fluency evidence document for this age band never mentions a rate
This is worth sitting with. The What Works Clearinghouse practice guide Providing Reading Interventions for Students in Grades 4–9 carries four recommendations, and the second is squarely about fluency with a Strong level of evidence behind it: "Provide purposeful fluency-building activities to help students read effortlessly."
So the federal guidance is not shy about fluency. What it declines to do is quantify it. In the guide's practice guide summary, the phrases "words per minute," "words correct," "speed" and "norms" appear not once.
Fluency is defined instead as "the ability to read text accurately, with ease, expression, and appropriate pacing," and the implementation advice is about purpose and prosody — "reread the same passage a total of 3–4 times, each time with a different purpose," and "focus some instructional time on reading with prosody."
The most-cited fluency table in the field and the strongest fluency evidence document for adolescent readers do not share a vocabulary. The table is a rate distribution; the guide is about accuracy, ease, expression and pacing, and nominates no number.
If your plan for a thirteen-year-old has become a words-per-minute goal with an intervention attached to it, that gap is worth noticing. Our comparison of reading intervention programs covers what the programs themselves claim to address, and our piece on structured literacy training programs covers what the training routes cost in hours.
What changed between 2006 and 2017, and where
The report devotes a section and two tables to this, and the pattern is more interesting than "scores went up."
The gains sit at the bottom of the distribution
Across all six grades the overall increase was 5 words correct per minute. But the increases are concentrated in the lower percentiles in the early grades.
At the 10th percentile in the winter of grade three, the figure moved from 36 in 2006 to 62 in 2017 — a 26-point increase, the largest in grades one to five. Grade two's 10th percentile rose by 12, 17 and 12 across fall, winter and spring. The 90th percentile, by contrast, gained an average of 4.
Averaged across all percentile levels, the report gives the gains as 7 words for grade one, 9 for grade two, 12 for grade three, 6 for grade four, 8 for grade five and 4 for grade six.
Grade six moved in both directions, and one cell went down
Grade six is the exception the report flags itself: eight of its fifteen figures rose, four fell by 1 to 4 words, and three were unchanged. The 75th percentile in spring fell from 177 to 173; the 50th percentile in spring fell from 150 to 146.
In grades one to five there is exactly one decrease: the grade two fall 50th percentile, from 51 to 50. And four cells across the whole table are identical in both years, including the grade four fall 50th percentile at 94.
One oddity the report does not comment on: in grade six, the 10th percentile is 89 in fall, 91 in winter and 91 in spring. At that percentile, in that grade, the compiled table shows essentially no movement across an entire school year.
Read Table 6 carefully
The report's Table 6 is headed "Average Differences in ORF Across PRs For Each Grade Level," with columns for Fall, Winter, Spring and an average. The seasonal columns hold figures like 41 and 30 for grade one, and 57 and 80 for grade three, which are far too large to be average changes in words per minute.
They are sums. For grade three, the five fall percentile differences in Table 5 are 6, 5, 12, 15 and 19, which total 57 — the figure in Table 6's fall column. The winter differences are 15, 17, 5, 17 and 26, totalling 80. The final "Ave" column then divides the three seasonal totals by fifteen to give 12, which matches the narrative. The arithmetic reproduces for every grade in the table.
So the only per-season figures in Table 6 are sums of five percentile differences, and reading them as average changes will overstate the shift by a factor of five. The narrative text, not the table, carries the averages.
The same paragraph also contains a small slip worth knowing about if you are checking the figures: it reports gains "in the 25th percentile and 9 WCPM in the 25th percentiles," where the second plainly means the 10th.
What that means for a child whose percentile you quoted from an old chart
A student at the 10th percentile on the 2006 winter grade three figures was reading 36 words correct per minute. The same reading rate against the 2017 figures sits well below the 10th percentile, which is 62. Nothing about the child changed. The comparison distribution did.
This is not a reason to prefer either table. It is a reason to name which table a percentile came from every single time you write one down — in a report, in a note, in an email to a school. A bare percentile with no table attached is not a finding.
How to read a words-correct-per-minute score in four questions
The questions below are the ones the documents above support asking. None of them requires a judgement the published record cannot bear.
1. Which table is this percentile from, and from which years of data?
Compiled 2017, compiled 2006, a publisher's own norms, or a local comparison. If the answer is "the fluency chart," that is not an answer yet. Write the source and the data years into the report alongside the number.
2. How many passages, and median, mean or single?
A single probe, a median of three, and the average of two are three different measurements with three different error profiles. Record which you did. If you are applying the 10-words-below-the-50th-percentile rule, the rule as published asks for the average of two unpracticed readings from grade-level materials, and if you gave one passage you have not met its stated condition.
3. Is this number a norm, a benchmark goal, or a cut point for risk?
A norm describes a distribution. A benchmark goal and a risk cut point predict an outcome against a named criterion and come with published classification accuracy. Only the second kind supports a sentence beginning "this score suggests he is on track for."
4. What is the confidence band, and does it cross a boundary?
Using the Oregon group's single-form figures, a grade three score sits inside an interval of roughly plus or minus thirty-two words correct per minute. Work out whether the interval crosses the band boundary you are about to report. If it does, report the band and the interval rather than the band alone.
There is a further twist worth knowing if your student is in the upper grades. The federal practice guide that recommends fluency-building for grades 4 through 9 never uses the phrase "words per minute" at all, and several published programs state their exit criteria in exactly that unit. That mismatch, and the two federal records behind it, are mapped in reading intervention for older students.
What this changes about a session and a progress report
Two practical consequences follow, and neither requires you to stop using the norms.
The first is about what you write down. A score, the number of passages, which score you took, the table you compared it against and the data years behind that table are five pieces of information, and four of them routinely go missing.
Our guidance on building a treatment plan and on running a first session both push toward recording the mechanics of a measurement rather than its headline. Fluency scores are the clearest case for it.
The second is about what you say out loud. AET's Code of Ethics, Revision #9, 2025, asks educational therapists under Instruction and Assessment Responsibilities and Competencies to "select and use appropriate assessment instruments, recognizing their limitations with respect to reliability, validity, and bias."
The compiled norms publish no reliability information and no standard error. Recognising that limitation is not optional under the clause, and the honest form of the sentence is the one that names the band rather than the point.
The same section asks therapists to "use only those assessment instruments for which they have been adequately trained," and to "seek interpretation of assessment data from professionals in related fields." A fluency probe is well inside most practitioners' training. The inference that a low fluency score indicates a specific disorder is not, and that boundary is the subject of our piece on what you may do with a psychoeducational report.
A note on citation form, since this site has used it: the Code numbers its provisions by section heading and letter rather than printing dotted references, so "Section Two, I.B" is a convention for pointing at a clause rather than a string you will find in the document.
What the norms are genuinely good for
Three uses are well supported by the documents above.
Flagging a student for closer attention. The report's own framing: students below benchmark "are good candidates for further diagnostic assessments to help teachers determine their skill strengths or weaknesses."
Sanity-checking a local expectation. This is the purpose the compilation was built for, and it works.
Describing a change over time in a student's own scores, with the error band attached, which is a within-child comparison and does not depend on the norm sample at all.
What they do not support is a target. The number in the 50th percentile column is a description of what children in three assessment samples between 2009 and 2014 did. Nothing in the report proposes it as the rate a particular child should be pulled toward, and reading fluency research has been clear for a long time that rate is one of three related components alongside accuracy and expression.
The companion report published alongside the norms, Tindal's Oral Reading Fluency: Outcomes from 30 Years of Research, quotes the National Reading Panel's definition — "the ability to read a text quickly, accurately, and with proper expression" — and the 2002 national assessment finding it cites puts all three in relation to comprehension.
If your intervention is raising a number and not the thing the number indicates, the number has stopped being an indicator. That caution belongs with the evidence discussion in our piece on whether educational therapy is evidence-based, and with the way reading program evidence ratings handle outcome measures that sit close to the trained task.
One more condition the table assumes
Every error estimate above was produced with an examiner in the room holding a stopwatch and a scoring protocol. A timed oral reading administered over a video call is a different measurement condition.
The publishers who permit remote administration say so in their own terms rather than in a blanket rule, which is the subject of our piece on what publishers permit for remote assessment. If you took the score over video, treat the standard errors in this piece as a floor rather than an estimate.
Where this sits alongside the rest of the assessment literature
Several of these documents share a vocabulary and mean different things by it, which is the pattern this site keeps finding:
Grade equivalent scores are a descriptive statistic that publishers themselves warn against reading as a target.
Outdated test norms turns on the fact that the ten-year rule appears in none of the documents cited for it.
Achievement test qualification levels found that the letter on a test is a purchasing rule for one storefront rather than an industry scale.
Reading program evidence ratings found five public systems rating programs against five different questions.
Reading comprehension assessment tools explains why two comprehension scores on the same student disagree.
The IDA Knowledge and Practice Standards traced what the word "aligned" is doing when a program claims it.
Math assessment tools for dyscalculia found the same qualification-letter inconsistency across publishers.
What "Orton-Gillingham based" actually means found a label no organisation adjudicates.
Compiled fluency norms belong in that family. The number is real, the method is published, and the word "norm" is carrying more weight in conversation than the document behind it ever claimed.
Where the published record runs out
Stating the gaps is more useful than papering over them, so here is what this piece could not establish.
Technical Report 1702 reports no standard error, reliability coefficient or confidence interval for any figure in its tables. The error figures above come from other publications by overlapping authors, using easyCBM and DIBELS passages, and they are not official error estimates for the compiled percentiles. No such estimate appears to be published.
The three source technical reports were not all re-read for this piece. The easyCBM 2014 norms report was read directly. The two DIBELS Data System percentile reports from 2011 were not, so the statement that their tables rest on different administration conventions rests on the 2013 administration paper rather than on those reports themselves.
AIMSweb is absent from the compilation, and the report says why. Pearson "refused to provide access to any of their data 'due to the changes in student data privacy laws nationwide,'" which the report attributes to a personal communication dated December 2016. Whatever the merits, the consequence is that a widely used CBM system contributed nothing to the compiled norms.
Two further data sets were obtained and deliberately excluded. CBMreading data from FastBridge Learning was left out because its scores "were significantly different from the scores from the other assessments we analyzed, perhaps due to the way in which their passages were constructed." Benchmark Assessor Live data from Read Naturally was left out because those scores "are most commonly collected only from students already identified as at-risk, vulnerable readers." Both exclusions are reasoned and published, and both are invisible in the table.
There is no compiled update after 2017. Searching the current published record turns up no fourth compilation, so the 2013–14 data remains the most recent input to the most recent compiled norms. If that changes, this piece is out of date.
Nothing here establishes what reading rates look like now. The honest statement is that the most recent compiled norms describe a decade-old distribution, not that children today read faster or slower than the table says.
None of this makes the compiled norms unusable. It makes them a description with a date and a method, which is what they always were, and which is a great deal more than most of the numbers on a progress report can claim.
Frequently asked questions
Are the Hasbrouck and Tindal oral reading fluency norms still current?
They are the most recent compiled oral reading fluency norms published, and no later compilation appears in the published record. The underlying data is older than the publication date: the three source data sets are from the 2009–10, 2010–11 and 2013–14 school years, so the newest input is more than a decade old. The report itself prints these years.
Does the 2017 update cover grades 7 and 8?
No. The 2017 compiled norms cover grades 1 through 6, with the grade one fall cell empty. The 2006 compiled norms covered mid-grade-one through the end of grade eight. For a seventh or eighth grader, the choices are the older compiled table, a publisher's criterion-referenced benchmarks such as the DIBELS 8th Edition goals that run through grade eight, or a local comparison.
How many passages should I give to get a usable score?
The publishers disagree, which is the honest answer. easyCBM's developers recommend a single probe; DIBELS Next's recommend three in succession with the median taken; and the interpretive rule published alongside the compiled chart asks for the average of two unpracticed readings. The Oregon group reports that adding one more form reduces the absolute standard error by roughly three words per minute. Whichever you choose, record it, because the three are not interchangeable.
Is the 50th percentile a reasonable goal for a student in intervention?
The report does not present it that way. It describes the compiled norms as published "to prevent a low-performing school or district from setting benchmark goals for their students at a level that was lower than it should be," which is a systems-level safeguard rather than an individual target.
A benchmark goal from a publisher's system is built to predict a later outcome against a named criterion and carries published classification accuracy; a percentile carries neither.
What is the standard error of a single oral reading fluency probe?
Published estimates vary by grade, passage construction and administration. Behavioral Research and Teaching reports predicted absolute standard errors for one form on one occasion of 10.32 at grade one rising to 16.23 at grade three. Christ and Silberglitt reported a median standard error of measurement of 10 words per minute across grades 1 to 5 under median-score administration, with a range of 4 to 15.
Poncy, Skinner and Axtell estimated 12 or 18 words for a single passage, falling to 5 or 7 with three.
Can a fluency score identify dyslexia or a reading disability?
No. A rate measure indicates how a student performed on a timed oral reading of one or more passages, which is why the compiled norms report frames below-benchmark performance as grounds for "further diagnostic assessments" rather than as a finding.
The National Center on Improving Literacy is a reasonable starting point for what screening is and is not designed to do, and what a screener can and cannot establish is covered in our piece on dyslexia screening tools. The scope question about interpreting assessment data sits with AET's Code of Ethics and the related-field referral it describes, and with our piece on what is educational therapy.
Join the Illuminate educator network
Illuminate Educational Therapy Group keeps a list for educational therapists, learning specialists and executive-function coaches. It is where we send our practitioner research as it publishes, and where we match therapists with students looking for one.
If assessment literacy is the part of the work you want to get sharper at, the pieces on reading comprehension assessment tools, written expression assessment tools and outdated test norms are the closest companions to this one. For the wider picture of how these roles differ, start with educational therapist versus learning specialist versus executive function coach.
Comments