Most advice about dyslexia-friendly typography gets repeated more confidently than it was ever measured. So before rewriting this guide I did the unglamorous part. I pulled the abstracts of the studies everyone cites, checked the sample sizes, and then set the same paragraph in four fonts in headless Chromium to see what the layout numbers actually do. Some of what this page used to say didn't survive. Why open with a correction? Because a confident wrong number does more damage on an accessibility page than a missing one. The 19.8 percent Lexend figure is the clearest casualty, and I'll get to it.
Yale's Center for Dyslexia and Creativity says dyslexia affects 20 percent of the population (Yale). That's Yale's figure, and I haven't checked it against other prevalence estimates. The point stands either way: a big slice of any audience meets your typography before it meets your argument. Clearing the WCAG contrast minimum is the floor, not the finish.
TL;DR: No dyslexia-specific font has shown a reliable gain in speed or accuracy in the studies checked below. Where one did (about 7 percent for Dyslexie, 39 readers), matching the spacing erased it. Spacing, line length and a layout that survives user overrides carry the weight.
What Does the Research on Dyslexia Fonts Actually Show?
Less than the product pages suggest, and what it does show points at spacing. Here's the evidence as I found it in the primary papers, with sample sizes, because "a study found" means nothing until you know whether it was 12 children or 1,200.
| Study | Who | What was compared | What it found |
|---|---|---|---|
| Wery and Diliberto, 2017 | 12 US students with dyslexia, grades 3 to 6 | OpenDyslexic vs Arial and Times New Roman | No gain on speed or accuracy; none of the students preferred it |
| Kuster et al., 2018 | 170 Dutch children (Exp. 1); 102 with and 45 without dyslexia (Exp. 2) | Dyslexie vs Arial, then vs Arial and Times New Roman | No difference in speed or accuracy; children mostly preferred Arial |
| Marinus et al., 2016 | 39 low-progress English readers | Dyslexie vs Arial, then Arial spaced like Dyslexie | About 7 percent faster than normal Arial; gone once spacing matched |
| Joseph and Powell, 2022 | 71 children aged 8 to 12 (37 with dyslexia) | Dyslexie vs Calibri, spacing held constant | Letters named 2.2 s faster; no effect on words, nonwords or passages |
| Zorzi et al., 2012 | 74 dyslexic children aged 8 to 14, plus a 20-child follow-up | Normal vs extra-wide letter spacing | Errors roughly halved; speed up about 0.3 syllables per second |
| Galliussi et al., 2020 | 128 children, 64 with dyslexia | Letter, word and font variations | Differences under 0.20 syllables per second; wide letters with default word gaps read worst |
Three things jump out when you read the papers instead of the headlines.
First, the font studies are blunt. Wery and Diliberto ran a single-subject design in one independent school, a careful method for a small group and a poor basis for sweeping claims. Kuster's team had the largest samples and still wrote that Dyslexie "neither benefits nor impedes" reading. Of the 102 children with dyslexia in their second experiment, 45.1 percent preferred Arial, 29.4 percent Times New Roman and 10.8 percent Dyslexie. Preference wasn't related to how well anyone read.
Second, the one positive font result has an asterisk the size of a billboard. Marinus and colleagues found Dyslexie readers got through about 7 percent more words per minute than with standard Arial. Then they spaced Arial the way Dyslexie is spaced, and the advantage stopped being significant. Sound familiar? A feature gets credited to the shape of the letters when the spacing did the work. Joseph and Powell, with spacing held fixed, found a 2.2-second faster letter-naming time and nothing on actual reading. They urge caution about their own result, because the two groups came from different schools.
Third, the spacing studies are better, not perfect. Zorzi's children read Times Roman at 14 pt with 2.5 pt added between letters, which is about 0.18em, half again as much as WCAG's 0.12em. Word and line spacing grew too in the first experiment, and the sentences were short and used common words. A replication with equal-length sentences, as Galliussi's team describes it, found fewer errors but no speed gain. Galliussi's own 128-child study then found the effects tiny, under 0.20 syllables per second, and that widening letters while leaving word gaps alone was the least readable combination. Spacing helps, but only in proportion. Wide letters crammed against default word gaps read worse than doing nothing.
Where is the evidence thinnest? Everywhere a web designer would want it. These are mostly children, mostly short passages or word lists, in a mix of English, Dutch, Italian and French. The font comparisons were run on Dyslexie and OpenDyslexic. I found no peer-reviewed trial of Lexend or Atkinson Hyperlegible at all.
That brings me to Lexend. This article used to say it "improved reading speed by 19.8% over Times New Roman in controlled testing" and cited Google Design. I fetched that page. It says Dr. Bonnie Shaver-Troup created Lexend, that she tested it with students, and that her doctoral dissertation at Azusa Pacific University concerned font size and spacing in second-grade reading outcomes. There's no 19.8 percent figure on it, no Times New Roman comparison, no sample size, and no claim of peer review. The number circulates in secondary write-ups that disagree with each other about how many children were tested. I couldn't trace it to a primary source, so it's gone. Lexend is still a fine, free, well-made font. It just doesn't come with proof.
My position, and you can disagree: treat the typeface as a preference setting and put the engineering effort into spacing and layout.
What Do the BDA and Dyslexia Scotland Guides Recommend?
Guidelines aren't studies, but they're what an accessibility reviewer will hold you to, so it helps to know what they say.
The British Dyslexia Association's 2023 style guide recommends sans serif fonts such as Arial, with Verdana, Tahoma, Century Gothic, Trebuchet, Calibri and Open Sans as alternatives, on the reasoning that letters look less crowded. It asks for 12 to 14 point text or the equivalent, such as 1 to 1.2em, notes that some readers will want larger, says to left-align without justification, and calls 1.5 line spacing preferable.
Dyslexia Scotland's page on typed formats is more cautious about the evidence and more specific about spacing. It says recent evidence shows no consistent advantage of sans serif over serif fonts, that specialised "dyslexia fonts" aren't reliably better than standard ones and should be used where an individual finds them helpful, and that letter, word and line spacing matter more than typeface for some readers. Its numbers: tracking of roughly +0.03em to +0.035em, word spacing at least three times the letter spacing, 1.5 line spacing, 50 to 75 characters per line with shorter lines suiting some readers, and at least 12pt. It also warns that excessive letter spacing can slow reading, which is Galliussi's result in a sentence.
So the two bodies disagree about serif versus sans and agree about everything that's actually CSS. I built the block in the next sections from the overlap: 0.03em letter spacing, 0.1em word spacing (3.3 times the letter value), 1.6 line-height, left-aligned text. Where does that leave a developer who just wants a sensible default? With the overlap, plus a switcher for the readers who want something else.
What Do These Fonts Look Like Next to Each Other?
The frame above sets one sentence in a system sans, Lexend, Atkinson Hyperlegible and OpenDyslexic, all running the CSS printed further down. Read the bdpq line before the paragraph. Those four letters are the mirror pairs that get cited in every dyslexia thread, and they're where the fonts disagree.
OpenDyslexic gives every letter a heavy bottom, on the theory that weight at the baseline stops a b flipping into a d. Whether that works is the question the studies above answer, and the answer is "not measurably". It also costs something the marketing skips. I measured each font with canvas measureText at 100px, and the same pangram sets 66 percent wider in OpenDyslexic than in Arial.
| Font | x-height at 100px | Pangram width | Width of "0" | Latin woff2 size |
|---|---|---|---|---|
| Arial (control) | 51.9 | 1945 | 55.6 | system |
| Lexend 400 | 52.5 | 2140 | 60.7 | 14,476 bytes |
| Atkinson Hyperlegible 400 | 49.6 | 1925 | 64.8 | 17,208 bytes |
| OpenDyslexic 400 | 56.0 | 3236 | 66.4 | 115,280 bytes |
Two readings. The x-heights sit within 6.4px of each other, so the "generous x-height" line you'll see in most write-ups isn't what separates these fonts. Glyph shapes are. In the files this page serves, Arial's capital I and lowercase l are the same bare stroke. Lexend gives the I slab serifs and the 1 a base. Atkinson Hyperlegible, designed by the Braille Institute in 2019 for low-vision readers, slashes the zero and tails the l, so Il1 and 0O stop being guesses (Braille Institute). It's also the only one of the three that stays as compact as Arial, 1 percent narrower on that pangram.
The last column is a real argument for opt-in. 115 KB for a font most readers don't need is eight times the weight of Lexend. Self-host whichever you pick, and load OpenDyslexic only when someone asks for it.
Notice one more column, though: the width of the zero. It's about to matter.
How Long Should a Line Be, Measured in ch?
Between 45 and 75 characters is the usual range, and Dyslexia Scotland's 50 to 75 matches it. Everyone writes max-width: 65ch and moves on. Is 65ch actually 65 characters?
No. ch is the advance width of the "0" glyph in the element's own font, and a zero is wider than an average lowercase letter. So I set the same four-paragraph passage at 16px, with this article's spacing, in a 65ch box in each font, and counted the characters on every full line using the browser's own line boxes.
| Font | Box at 65ch | Characters per line at 65ch | Measure aimed at ~60 | Characters per line | Under a 1.4.12 override |
|---|---|---|---|---|---|
| Arial | 578px | 72.7 | 55ch | 61.0 | 49.3 |
| Lexend | 631px | 72.7 | 55ch | 61.0 | 50.9 |
| Atkinson Hyperlegible | 674px | 85.7 | 46ch | 60.0 | 48.7 |
| OpenDyslexic | 691px | 53.5 | 73ch | 60.5 | 53.2 |
At 65ch, Atkinson runs 85.7 characters per line, past the top of the range, while OpenDyslexic gives 53.5. One value can't serve every font. If you ship a font switcher, the measure has to travel with the font-family, which is why the CSS below stores it in a custom property next to the typeface. And if you set the measure in pixels instead, swapping to OpenDyslexic quietly turns a 60-character line into a much shorter one.
Here's the block the frame above runs, verbatim.
.article-body {
--measure: 55ch;
max-width: min(100%, var(--measure));
font-size: clamp(1rem, 0.9rem + 0.5vw, 1.25rem);
line-height: 1.6;
letter-spacing: 0.03em;
word-spacing: 0.1em;
text-align: left;
overflow-wrap: break-word;
}
.system { font-family: system-ui, -apple-system, sans-serif; }
.lexend { font-family: "Lexend", system-ui, sans-serif; }
.atkinson { font-family: "Atkinson Hyperlegible", system-ui, sans-serif; --measure: 46ch; }
.dyslexic { font-family: "OpenDyslexic", system-ui, sans-serif; --measure: 73ch; }
Why clamp(1rem, 0.9rem + 0.5vw, 1.25rem)? The rem floor means the text starts at whatever size the reader set in their browser, 16px by default, and never goes under it. The vw term makes it grow with the window: the arithmetic says the floor holds until the viewport is 320px wide (0.9 x 16 + 0.005 x 320 = 16) and the 20px ceiling arrives at 1120px. In between, text glides. The min(100%, ...) keeps a 73ch OpenDyslexic column from overflowing a phone. And text-align: left follows both guides' advice against justification, because justified text opens uneven gaps.
A dyslexia font is a preference. Spacing, line length and a layout that survives overrides are the actual work.
What Does WCAG 1.4.12 Require, and What Survives It?
Success Criterion 1.4.12 Text Spacing (Level AA) says that when a user sets four properties and changes nothing else, no content or functionality may be lost (W3C). The four values are exact, and each is a multiple of the font size.
| Property | Minimum the page must tolerate | At 16px | At 18px | At 20px |
|---|---|---|---|---|
| Line height | 1.5 x font size | 24px | 27px | 30px |
| Paragraph spacing | 2 x font size | 32px | 36px | 40px |
| Letter spacing | 0.12 x font size | 1.92px | 2.16px | 2.4px |
| Word spacing | 0.16 x font size | 2.56px | 2.88px | 3.2px |
I ran that arithmetic in node rather than trusting my head, and the output matches the table. Note what the criterion doesn't say: your default design needn't use these values. It says your page can't break when someone else does. The values come from a study by McLeish that tested letter spacing from 0.04 to 0.25em, with reading speed rising and then flattening near 0.20em, and the W3C calls them a baseline, not a ceiling. A language or script that doesn't use one of the properties can conform with only the ones that apply.
"No loss" means two concrete failures: text cut off by its box, and text overlapping other content. Truncating with an ellipsis passes only if the reader can reveal the rest. The failure technique is F104, and the sufficient technique is C36, allowing for text spacing override.
So what does breakage look like? I built the two usual suspects and applied the override with the same declarations a bookmarklet uses: line-height: 1.5, letter-spacing: 0.12em and word-spacing: 0.16em, all !important on every element, plus p { margin-bottom: 2em }. Measured in the frame's 720px layout, the fixed-height card's content needed 130px inside a 104px box, so its last line (the word "page.") was cut off. The fixed-height button needed 46px inside 38px. The min-height versions grew to 130px and 62px instead and lost nothing.
.card--fixed { height: 5.5rem; overflow: hidden; }
.card--flex { min-height: 5.5rem; }
.btn--fixed { height: 2.5rem; overflow: hidden; }
.btn--flex { min-height: 2.75rem; padding-block: 0.5rem; }
Tick the override box in the frame and you can watch it happen. The rule behind it is short. Text containers get min-height, never height. Buttons get padding instead of a fixed height. overflow: hidden on anything that holds words is a bug waiting for a user stylesheet, and -webkit-line-clamp has the same problem unless a click reveals the full text.
What Breaks at 200 Percent Zoom?
Three different criteria get blurred together here, so separate them.
SC 1.4.4 Resize Text (AA) says text can be resized up to 200 percent without assistive technology, with no loss of content or functionality (W3C). Its failure F94 names the usual culprit: text sized in viewport units can't be resized by zooming, because the viewport shrinks as you zoom and the unit shrinks with it. SC 1.4.10 Reflow (AA) says content must work at 320 CSS pixels wide without scrolling in two directions, which is what a 1280px window looks like at 400 percent zoom (W3C).
Does clamp() with a vw term break 1.4.4? That depends on the numbers. I worked it out in node for our clamp on a 1280px window, where zoom z shrinks the CSS viewport to 1280/z and the displayed size is z times the computed size. I included a pure-vw size for contrast.
| Zoom | CSS viewport | Our clamp, displayed | Growth vs 100% | Pure vw, growth |
|---|---|---|---|---|
| 100% | 1280px | 20.0px | 1.00x | 1.00x |
| 150% | 853px | 28.0px | 1.40x | 1.00x |
| 200% | 640px | 35.2px | 1.76x | 1.00x |
| 300% | 427px | 49.6px | 2.48x | 1.00x |
| 500% | 256px | 80.0px | 4.00x | 1.00x |
Pure viewport units never grow, which is F94 in a column of ones. Our clamp does grow, but look at the 200 percent row: 1.76x, not 2x. It reaches exactly 2x at about 206 percent zoom, where 14.4z + 6.4 = 40. A strict auditor measuring at precisely 200 percent could flag that. Maxwell Barvian's analysis in Smashing Magazine gives a simple rule of thumb, that the maximum should be no more than 2.5 times the minimum, so that text at 500 percent zoom is still at least double its 100 percent size (Smashing Magazine). Ours is 1.25 times, so it passes that test by a wide margin. If you'd rather not argue with an auditor, shrink the vw term or drop it and use plain rem. My view: a four-pixel fluid range isn't worth a debate about 1.4.4, so on body text I'd keep it small or fixed and save fluid scaling for headings. Our fluid typography guide covers headings properly.
Now the failures that bite in practice, in rough order of how often they'll hit you:
- Fixed-height containers. Covered above. They fail at override and again at zoom.
- The measure shrinks under override.
chdoesn't grow with letter-spacing, so the same box holds fewer characters. In the table earlier, Arial drops from 61.0 to 49.3 characters per line and OpenDyslexic from 60.5 to 53.2. Still inside 45 to 75, which is why I aimed at 60 instead of 75 in the first place. - Sticky headers at 400 percent. A 64px header on a 900px-tall window is 7 percent of the screen. At 400 percent zoom the window is 225 CSS pixels tall and the same header eats 28 percent of it. Let the header scroll away below a width, or shrink it.
- Justified or hyphenated text. Justification gets worse as columns narrow.
- Absolute-positioned labels. Anything placed relative to text that grows will collide with it.
What About Color, Contrast, and Background Choices?
Pure white behind pure black passes at the maximum ratio and still gets reported as harsh by some dyslexic readers. Dyslexia Scotland cites studies where off-white or tinted backgrounds with dark text improved reading speed over stark white, and says in the same breath that these adjustments don't help every dyslexic reader and that preferences vary. Read that as "give people the option", not "everyone needs cream". Does a softer background cost you anything? Only if you let the contrast ratio slide with it.
If you do soften the page, check the numbers. I computed WCAG contrast ratios for a few pairs: #000000 on #ffffff is 21.00:1, #1a1a1a on #faf7f2 is 16.29:1, #333333 on #faf7f2 is 11.82:1, and a mid-grey #595959 on the same cream is 6.55:1, still over the 4.5:1 floor for body text. You can drop the glare a long way and keep a huge margin. Bake the pair into your tokens instead of re-deciding it on every page, which is the argument in our accessibility tokens guide.
How Do You Audit an Existing Site for This?
Run the body text through the contrast checker on this page first. Then work down a short list, using a real browser rather than a mockup. Which items fail first? In the test above it was the containers that broke, never the font.
- Count characters per line at your widest and narrowest layouts, in the font you actually ship. Not your design width. If it's over 75 on desktop, that's a bigger problem than any font debate. Our line length and vertical rhythm guide goes deeper on measure.
- Set line-height to 1.5 or more, letter-spacing near 0.03em, and word-spacing about three times that.
- Apply the four 1.4.12 values with a bookmarklet or user stylesheet and look for clipped or overlapping text. Check buttons, cards, table headers and form labels first.
- Zoom to 200 percent, then 400 percent at 1280px wide, and confirm there's no sideways scroll and the sticky header leaves room to read.
- Check body text is in
rem, neverpx, and that nothing relies on a barevwfor font size. - If you offer a font switcher, store the measure with the font and rerun the characters-per-line count for each one.
- Recheck every text pair in dark mode, because dark-mode typography fails contrast in ways a light-mode audit never surfaces. For the wider product, the UX audit checklist gives the order to work in.
None of this is a font swap you do once. It's a handful of habits (spacing, measure, units, containers that can grow) applied consistently, and they help nearly everyone who reads your pages, which beats a single typeface the evidence can't back.
Check a pair before you argue about it
Contrast is the one accessibility claim you can settle in four seconds, and it's still the one most often settled by opinion. Black on white scores 21:1 and clears every threshold here. Whether it's comfortable for twenty minutes of reading is a separate question, and the answer varies by reader. Pass first, then tune.