The Measure: Line Length, Leading, and What Is Actually Measured

The most quoted sentence in modern typography contains a hedge that almost nobody repeats. Here is Robert Bringhurst, as reproduced in The Elements of Typographic Style Applied to the Web:

Anything from 45 to 75 characters is widely regarded as a satisfactory length of line for a single-column page set in a serifed text face in a text size. The 66-character line (counting both letters and spaces) is widely regarded as ideal. For multiple column work, a better average is 40 to 50 characters.

"Widely regarded." Twice. Bringhurst is reporting a consensus among typographers, accurately and carefully, and he does not claim it is an experimental result. The design industry dropped the hedge, and the range has since been quoted as though a laboratory produced it. It did not. When the laboratory did get involved, it produced something considerably messier, and in places it produced the opposite answer.

What the experiments actually found

Mary Dyson, of the Department of Typography and Graphic Communication at Reading, reviewed the screen-reading literature in "How physical text layout affects reading from screen". Her summary of the line length work is not what a designer expects:

Although longer line lengths may be read faster, people prefer a more moderate length.

The studies she assembles are consistent about the first half of that sentence and consistent about the second. Dyson and Kipping (1998a) found "the longest line read faster than the shortest line tested", testing 100 characters per line against 25. Duchnicky and Kolers (1983) found that "the more characters per line, the faster the reading rate". Youngman and Scharff (1998) reported that "the fastest reaction time was with the longest line length".

Not 66. Not 75. One hundred characters per line, read faster than any shorter setting tested.

Here is the set of results, with the usual caution that these are different tasks, materials, screens and decades, which is exactly the problem:

The Tinker and Paterson line is worth pausing on, because it is the one that sounds like the received wisdom, and it is from 1929, on paper, at 10 point. Three inches of 10 point type is in the region Bringhurst describes. The screen studies from sixty years later point the other way. That is not a contradiction anyone has resolved; it is two different reading situations, and the convention was built on the first one.

The divergence, stated plainly

Dyson's review gives the cleanest example of the split. Of her own 1998 study: "A medium line length of 55 characters per line was rated as easiest to read, but was not read the fastest." Of Youngman and Scharff: "the longest line length was disliked the most, despite good performance." On columns, the same pattern: "With columns, a single wide column is read faster, but narrow multiple columns are preferred."

So the reader is a poor witness to their own reading rate. That is not a throwaway observation, and it has since been measured directly in a different domain. Wallace and colleagues, in ACM Transactions on Computer-Human Interaction in 2022, ran 352 remote participants through a font preference tournament and a separate reading-speed test. Their participants read "35% faster in their fastest font compared to their slowest font (232 WPM, on average)", and 76% of them showed a large effect of font on reading speed. And then this:

Participants read the fastest in their most preferred font 20% of the time, but they also read the slowest in their preferred font 20% of the time, which is precisely at chance level.

Five fonts, twenty per cent each way. Preference carried no information about speed at all. The study is about fonts rather than line length, so I am extending it, and you should hold that extension loosely. But it makes the line length pattern less surprising: if readers cannot tell which typeface they read fastest in, there is no reason to expect them to know which measure they read fastest in either.

Which leaves an awkward design question that the research does not answer. If long lines are faster and short lines are preferred, which do you serve? Reading speed is easy to measure and rarely what anyone is actually optimising. Whether a reader stays with a page for twenty minutes is hard to measure and usually the whole point. I have not found a study that measures sustained voluntary reading against measure, and I would be glad to be sent one.

Leading, where the evidence is thinner than the advice

The typographic rule is that longer lines need more leading, because the return sweep has further to travel and needs a bigger target. Dyson states the rule and then states its evidential position:

Increasing interlinear space might help in more accurate return sweeps, but this has not been explored and limited studies from the early 1980s are inconclusive.

Those studies: Kolers and colleagues (1981) found double spacing only "marginally superior" to single; Kruk and Muter (1984) found "reading was significantly slower with single spacing compared to double spacing". A marginal effect and a significant one, in the same decade, on the same question.

The firmest number available is not a reading-speed finding at all, it is an accessibility floor. WCAG Success Criterion 1.4.12 Text Spacing requires that no loss of content or functionality occurs when a user sets, among other things, "Line height (line spacing) to at least 1.5 times the font size" and "Spacing following paragraphs to at least 2 times the font size", with letter spacing at 0.12 times the font size and word spacing at 0.16. The Understanding document is explicit that this is about people with low vision and dyslexia, noting that "People with dyslexia may increase space between lines, words, and letters to increase reading speed."

That is a requirement that your layout must survive 1.5, not a claim that 1.5 is optimal. It is still the most defensible number in the neighbourhood, because it is the one you are obliged to support.

The ch unit is not counting characters

Here is the mistake I made for years, in production, on sites that are still live. The CSS looks like a direct implementation of Bringhurst:

article { max-width: 66ch;
}

article { max-width: 66ch;
}

article { max-width: 66ch;
}

But ch does not mean "character". MDN's definition: it "Represents the width or, more precisely, the advance measure of the glyph 0 (zero, the Unicode character U+0030) in the element's font", and "In cases where determining the measure of the 0 glyph is impossible or impractical, it must be assumed to be 0.5em wide by 1em tall."

In most text faces the zero is a lining figure, cut to align with the capitals and to tabulate with the other digits. It is wide. Running prose, meanwhile, is mostly lowercase, and roughly one character in six is a word space, which is typically around a quarter of an em. So the average advance of a character in a paragraph is meaningfully narrower than the advance of a zero, and 66ch therefore sets a measure that holds noticeably more than 66 characters of prose. How much more depends entirely on the face, and I have not measured it across a useful sample, so I will not give you a multiplier. The direction is not in doubt.

If you have been setting 65ch or 70ch and feeling that your columns run slightly long, this is why. The remedy is not a cleverer unit. It is to count.

A defensible way to set a measure

Given a literature that does not agree with itself, "defensible" cannot mean "optimal". It means you can say what you did and why, and the reasoning survives someone reading the papers.

Four steps.

1. Count characters in the actual face, at the actual size, with the actual copy

Set a real paragraph, not lorem ipsum, in the font you are shipping. Count the characters on five or six full lines and take the mean. That number is your measure. Everything else is a proxy for it, and every proxy is face-dependent.

2. Pick the target from the situation, not from the range

The experimental picture supports a wider setting than the convention when raw reading rate is the goal, and the preference picture supports a narrower one. So decide which you are serving. Long-form editorial where the reader has chosen to be there: favour the preferred end, because the risk is abandonment, not slowness. Reference material that people scan under time pressure: the evidence for longer lines is better than the convention admits, and Dyson's 100-character result should make you less afraid of a wide column than you probably are.

3. Set leading against the measure, and make 1.5 the floor for body text

WCAG obliges your layout to survive 1.5. Setting body text at or above 1.5 by default means you are not relying on a fragile override. Then adjust upward as the measure widens, on the return-sweep argument, while knowing that the argument is typographic tradition with inconclusive experimental support rather than a finding.

4. Express it in CSS so it degrades sensibly

:root {
  /* measured by counting, not assumed from ch */
  --measure: 34rem;
  /* approx 68 characters in the shipping face */
}

article {
  max-width: var(--measure);
  font-size: 1.125rem;
  line-height: 1.55;
  /* at or above the WCAG 1.4.12 floor of 1.5 */
}

/* wider column, more leading: tradition, not evidence */
@media (min-width: 60rem) {
  article {
    max-width: 42rem;
    line-height: 1.65;
  }
}

:root {
  /* measured by counting, not assumed from ch */
  --measure: 34rem;
  /* approx 68 characters in the shipping face */
}

article {
  max-width: var(--measure);
  font-size: 1.125rem;
  line-height: 1.55;
  /* at or above the WCAG 1.4.12 floor of 1.5 */
}

/* wider column, more leading: tradition, not evidence */
@media (min-width: 60rem) {
  article {
    max-width: 42rem;
    line-height: 1.65;
  }
}

:root {
  /* measured by counting, not assumed from ch */
  --measure: 34rem;
  /* approx 68 characters in the shipping face */
}

article {
  max-width: var(--measure);
  font-size: 1.125rem;
  line-height: 1.55;
  /* at or above the WCAG 1.4.12 floor of 1.5 */
}

/* wider column, more leading: tradition, not evidence */
@media (min-width: 60rem) {
  article {
    max-width: 42rem;
    line-height: 1.65;
  }
}

Note the unit. rem is tied to the root font size, which MDN notes is commonly 16px but which "user-defined preferences may modify". A measure set in rem scales with a reader who has turned their text up; a measure set in px does not, and turns into a very long line at very large text. That is a real accessibility difference and it costs nothing.

What I got wrong, and what is still open

Three admissions.

I treated 45 to 75 as an empirical finding and cited it to clients as though it were one. It is a report of professional consensus, made by someone who was careful to say so. The consensus may well be right. It is not evidence, and presenting it as evidence was sloppy.

I used ch as a character count. It is the width of a zero.

And I assumed that what readers say is comfortable tracks what actually works for them. The line length literature says it does not, at least for speed, and the Wallace study says the same thing about fonts at exactly chance level.

What remains genuinely unknown, as far as I can establish: nobody has resolved why the screen studies and the print convention point in different directions, and Dyson's own diagnosis is that they may not be comparable at all, since the differences between studies "include the measures, materials, and tasks used and participants' characteristics" and technology itself changed underneath the work. Her recommendation for future research is characteristically modest and, twenty-odd years on, still largely undone: find "the line length at which an increase in characters per line slows down reading". Until someone does, the honest position is that the measure is a judgement, not a calculation, and that anyone quoting a single number at you, including me, is quoting a convention.

Sources

• Robert Bringhurst, The Elements of Typographic Style, quoted in Richard Rutter, "Choose a comfortable measure", The Elements of Typographic Style Applied to the Web, section 2.1.2. The 45 to 75 range, the 66-character line, and the 40 to 50 figure for multiple columns. https://webtypography.net/2.1.2

• Mary C. Dyson, "How physical text layout affects reading from screen", Department of Typography and Graphic Communication, University of Reading. Line length and reading speed results from Dyson and Kipping (1998a), Duchnicky and Kolers (1983) and Youngman and Scharff (1998); the speed versus preference divergence; the interlinear spacing evidence from Kolers et al. (1981) and Kruk and Muter (1984); the single-column result; and the statement of methodological differences between studies. https://stu.westga.edu/~ssynan1/literacy/Dyson.pdf

• Mary C. Dyson and Mark Haselgrove, "The influence of reading speed and line length on the effectiveness of reading from screen", International Journal of Human-Computer Studies, volume 54, issue 4, 2001, pages 585-612, doi 10.1006/ijhc.2001.0458. The finding that a medium line length of 55 characters per line supports effective reading at normal and fast speeds. https://www.sciencedirect.com/science/article/abs/pii/S1071581901904586

• Michael L. Bernard, Marissa Fernandez, Spring Hull and Barbara S. Chaparro, "The Effects of Line Length on Children and Adults' Perceived and Actual Online Reading Performance", Proceedings of the Human Factors and Ergonomics Society Annual Meeting, volume 47, issue 11, October 2003, pages 1375-1379. No differences in reading time or efficiency; adults preferred shorter line lengths. https://journals.sagepub.com/doi/10.1177/154193120304701112

• Human Factors International, "Optimal Line Length", newsletter review. Summaries of Tinker and Paterson (1929), Dyson and Kipping (1998), Youngman and Scharff (1999) and Bernard, Fernandez and Hull (2002), including the inch measurements and the preference findings. https://www.humanfactors.com/newsletters/optimal_line_length.html

• Shaun Wallace, Zoya Bylinskii, Jonathan Dobres, Bernard Kerr, Sam Berlow, Rick Treitman, Nirmal Kumawat, Kathleen Arpin, Dave B. Miller, Jeff Huang and Ben D. Sawyer, "Towards Individuated Reading Experiences: Different Fonts Increase Reading Speed for Different Individuals", ACM Transactions on Computer-Human Interaction, volume 29, issue 4, article 38, 2022, doi 10.1145/3502222. 352 participants; 35% difference between fastest and slowest font; preference predicting speed at chance level. https://jeffhuang.com/papers/Readability_TOCHI22.pdf

• W3C Web Accessibility Initiative, "Understanding Success Criterion 1.4.12: Text Spacing", WCAG 2.2. Line height at least 1.5 times the font size, paragraph spacing at least 2 times, letter spacing 0.12, word spacing 0.16, and the benefit statements for low vision and dyslexia. https://www.w3.org/WAI/WCAG22/Understanding/text-spacing.html

• MDN Web Docs, "length", CSS values and units. Definitions of the ch unit as the advance measure of the glyph 0 with a 0.5em fallback, and of rem as the root font size subject to user preference. https://developer.mozilla.org/en-US/docs/Web/CSS/length

Related reading

• Variable Fonts: The Real Production Trade

• Colour Spaces, and Why Your Gradient Goes Grey

• Hierarchy With One Typeface

• A Type Scale Is a Convention, Not a Law

More from the Moc Mallen Studio journal