A single number for how hard a text is to read
Rudolf Flesch published the Reading Ease formula in 1948 in the Journal of Applied Psychology, fitting three constants against comprehension test results so that the output landed roughly between 0 and 100 with higher meaning easier. It has outlived almost every competitor for one reason: it needs only counts, so anyone can compute it by hand and nobody can argue about the inputs.
The score is built from two averages. Average sentence length is subtracted with a small weight of 1.015, so each extra word per sentence costs about a point. Syllables per word is subtracted with a very large weight of 84.6, so each extra tenth of a syllable per word costs 8.46 points. The starting constant of 206.835 exists only to place a plain, short-sentence English text near the top of the range.
That weighting is the formula's personality. Reading Ease punishes long words far harder than the companion Flesch-Kincaid grade level does relative to sentence length. A document written in short sentences of Latinate vocabulary scores badly here and only moderately badly on the grade scale. If you have a legal or regulatory target expressed as a Flesch score, vocabulary is where you will find the points.
The scale is not bounded by the arithmetic. Very short sentences made entirely of monosyllables produce scores above 100, and pages of 50-word sentences full of four-syllable words produce negative scores. Both are legitimate outputs; they simply mean 'past the end of the useful range'.
The bands, and where they carry legal weight
Flesch attached descriptive labels to ten-point bands, and those labels are still what practitioners quote. Ninety to 100 is very easy, readable by an average eleven-year-old. Sixty to 70 is plain English, the target for general-audience writing. Thirty to 50 is difficult, typical of academic prose. Below 30 is very difficult, the territory of contracts and technical standards.
What makes Reading Ease unusual among readability formulas is that it appears in law. Following the National Association of Insurance Commissioners' Life and Health Insurance Policy Language Simplification Model Act, many US states require insurance policy forms to reach a minimum Flesch Reading Ease score before they may be issued, commonly 40 or 45. The precise figure, the sampling method and the exemptions are set by each state's insurance code, so a drafter working across states has to check the requirement for each jurisdiction rather than assume a national number.
Because a legal minimum makes the score a pass mark rather than a guide, the target inversion in this calculator matters. Fix the syllables per word you have and solve for the sentence length the target allows: ASL = (206.835 − target − 84.6·ASW) ÷ 1.015. At 1.5 syllables per word a score of 45 permits sentences averaging 34.4 words — generous. At 1.9 syllables per word the same target permits only 1.1 words per sentence, which is to say it is unreachable. Vocabulary sets a ceiling on the score that no amount of sentence surgery can lift.
Worked example: a policy paragraph at 1.8 syllables per word
You have a 200-word sample of policy wording. It contains 10 sentences and 360 syllables. Your state requires a minimum Flesch Reading Ease score of 45.
- Average sentence length. 200 ÷ 10 = 20.00 words per sentence.
- Syllables per word. 360 ÷ 200 = 1.800.
- Sentence-length penalty. 1.015 × 20.00 = 20.30 points.
- Word-length penalty. 84.6 × 1.800 = 152.28 points.
- Score. 206.835 − 20.30 − 152.28 = 34.26. That is in the 'difficult' band and below the 45 required.
- Sentence length that would reach 45. (206.835 − 45 − 152.28) ÷ 1.015 = 9.555 ÷ 1.015 = 9.41 words per sentence. At 200 words that means 200 ÷ 9.41 = 21.3 sentences instead of 10.
- Vocabulary route instead. Hold sentence length at 20 and solve for syllables per word: (206.835 − 45 − 20.30) ÷ 84.6 = 141.535 ÷ 84.6 = 1.673 syllables per word. Dropping from 1.800 to 1.673 is about 25 syllables removed from the 200-word sample.
Compare the two routes. Halving every sentence is drastic and often destroys the conditional structure that policy language needs. Removing 25 syllables — turning 'prior to the commencement of' into 'before', 'in the event that' into 'if', 'utilisation' into 'use' — is 25 small substitutions and leaves the legal structure intact. On a Flesch target, vocabulary is nearly always the cheaper lever, and that is a direct consequence of the 84.6 coefficient.
How to act on the score
Decide first whether you are meeting a requirement or improving a draft. If a regulator or a house style sets a minimum, the score is a pass mark and the target inversion tells you which lever to pull and by how much. If you are simply writing better, the score is most useful as a before-and-after measure on the same document: a rise of ten or fifteen points between drafts is a real change, while any single absolute value is soft.
Sample properly. Three separate passages of 200 to 300 words from different parts of a document, averaged, will be far more representative than one passage of 100 words. Never include headings, bullet fragments, tables, addresses or reference lists — none is a sentence, and all of them wreck both averages. Several state insurance codes prescribe their own sampling method precisely because the number is so sensitive to what you include.
Watch two failure modes. The first is gaming: chopping every sentence at its conjunctions raises the score and can make text harder to follow, because the connectives that showed how clauses relate have gone. The second is false comfort: the formula cannot see whether a short word is a term of art. 'The insured shall have no right of subrogation' is mostly short words and scores respectably while remaining opaque to anyone who does not know what subrogation is. Reading Ease measures form, never familiarity.
Where the score is high and readers still struggle, the problem is usually organisation, unexplained terminology, or an unstated logical structure — none of which any formula detects. The productive next step there is a reader test, not another metric. If you want the same measurements expressed as a school grade instead of a 0-to-100 scale, the Flesch-Kincaid grade level calculator uses the identical inputs with different coefficients.
Reading Ease bands and what falls in them
| Score | Band | Approximate school level | Typical material |
|---|---|---|---|
| 90-100 | Very easy | 5th grade | Children's books, simple public notices |
| 80-90 | Easy | 6th grade | Popular fiction, consumer instructions |
| 70-80 | Fairly easy | 7th grade | Magazine features, plain-language guidance |
| 60-70 | Plain English | 8th to 9th grade | General newspapers, most web copy |
| 50-60 | Fairly difficult | 10th to 12th grade | Quality broadsheet analysis, trade press |
| 30-50 | Difficult | College | Academic prose, technical documentation |
| 0-30 | Very difficult | College graduate | Contracts, standards, primary legal texts |
The bands are Flesch's own. Scores above 100 and below 0 are arithmetically possible and simply mean the text sits beyond the range the scale was designed for.
What distorts a Flesch score
- Counting headings as sentences. A three-word heading with no full stop is not a sentence, and including it drags average sentence length down and the score artificially up.
- Including reference lists and tables. Citation strings and numeric cells are not prose and produce meaningless syllable and sentence counts.
- Sampling under 100 words. One 45-word sentence in a 100-word sample moves the score by more than ten points. Use 200 to 300 words per sample and average three samples.
- Assuming software agrees. Word, online checkers and hand counting differ on abbreviations, hyphenated words and syllable heuristics, typically by two to four points. State which tool produced a score when it is a compliance figure.
- Splitting sentences to game the number. Removing conjunctions raises the score and can lower comprehension, because the reader now has to reconstruct the logical relations you deleted.
- Believing short words are known words. Terms of art score as easy. The formula cannot model vocabulary familiarity at all; Dale-Chall is the formula that tries to.
- Applying it to non-English text. The constants were fitted on English. Adapted versions exist for other languages with different coefficients, and using the English formula on translated text produces a number with no meaning.
How it compares with the other readability formulas
Reading Ease and Flesch-Kincaid Grade Level are siblings, built from the same two averages. Reading Ease came first, in 1948; Kincaid and colleagues refitted the coefficients in 1975 for the US Navy and rescaled the output to school grades. They will always move in opposite directions, since one measures ease and the other difficulty, but they are not simple inverses of each other because the relative weight on vocabulary differs — Reading Ease penalises long words much more aggressively.
Among the alternatives, Gunning Fog counts words of three or more syllables rather than averaging syllables across all words. SMOG counts polysyllabic words in a fixed 30-sentence sample and was calibrated against full comprehension rather than the 50 per cent criterion behind the Flesch formulas, which is why health communicators often prefer it for consent forms and dosing instructions. Dale-Chall takes a different approach entirely, counting words absent from a list of about 3,000 that fourth graders reliably recognise, making it the only widely used formula that models familiarity rather than length. Expect two to three grades of disagreement between formulas on the same text; that spread is a property of the formulas, not an error.
In everyday drafting the practical pairing is this score with the Flesch-Kincaid grade level, because regulators speak in one and educators in the other. Once the text is settled, the reading time calculator turns the word count into minutes for the reader, the words to pages calculator converts a page requirement into the word budget you actually have to write, and the speech time calculator handles the case where the text will be heard rather than read — a listener cannot re-read a 40-word sentence, so spoken material needs a higher Reading Ease score than the same content on the page.