Pick a serif face designed for text, set it around 10 to 12 point, and give it generous line spacing. That covers most of the decision. The uncomfortable finding in the research is that beyond obvious extremes, the choice of typeface matters far less than size and spacing, and the confident advice you will read about serifs is not supported by evidence.
What Actually Makes a Face Readable
Three properties recur across a century of investigation, and they are about proportion rather than style.
Apparent size, meaning x-height. The height of the lowercase letters matters more than the nominal point size. A committee of the British Association for the Advancement of Science put it plainly in 1913: "Legibility depends mainly on the height and breadth of the short letters." Two faces at the same point size can differ substantially in apparent size, which is why comparing "11 point to 11 point" is misleading.
Moderate stroke contrast. Very fine hairlines disappear, very heavy strokes close up the white space inside letters. The 1913 report: "The contrast between the finer and the heavier strokes should not be great, for hair-strokes are difficult to see. On the other hand, a very heavy-faced type suffers in legibility through diminution of the white inter-spaces."
Modern vision science agrees from the opposite direction. A 2020 review of low-vision reading found reading speed "was optimal for the standard Courier font, and declined for the very thin or the very thick stroke-widths". Both extremes hurt.
Performance in mass, not as a specimen. A face that looks striking in a line can be punishing over 300 pages. The type founder Theodore De Vinne, writing in 1900: "The attractiveness of a very black-faced type when used in one line or in a few lines becomes repelling when it is used in a mass."
Edmund Gress reported the same effect from an actual test in 1917, where a sans-serif face won as isolated lines and lost as a page: "it is not as legible in the mass as in single lines."
The Serif Question, Honestly
Serif faces dominate book interiors. That is a fact about publishing, and a survey of contemporary books and newspapers found the text faces "all of which are seriffed romans".
What no one has shown is that serifs make text easier to read.
A 2005 vision-science study isolated the variable properly, varying only serif size while holding the letterforms constant, and reported "no difference in legibility between typefaces that differ only in the presence or absence of serifs". It found a hint that at threshold sizes serifs "may actually interfere ever so slightly". A 2022 study comparing Times and Helvetica at matched x-height concluded their legibility "is similar when investigated under equivalent conditions". A review of low-vision reading states flatly: "The presence of serifs is not a limiting factor on reading speed."
The UK Association for Accessible Formats says the same in its 2024 guidance: "there is no definitive evidence as to whether serif or sans serif typefaces are easier to read for most people, and some evidence that it makes no difference."
None of this is new. A 1926 Medical Research Council report surveying a century of legibility research concluded: "it is worth reiterating how extraordinarily little solid agreement is to be found over the whole field of typographical investigation."
So choose a serif face for a novel because it is what readers expect of a book and because the tradition is deep, not because it is measurably easier to read. Both are legitimate reasons. Only one of them is true.
Real Numbers
A 2011 study measured 200 novels published since 2000 with a loupe. Text in both hardcovers and paperbacks averaged an x-height of 1.4 to 1.9 mm, and the book faces sampled had x-heights of 0.42 to 0.43 of the body size. Those two figures together imply a body size of roughly 10 to 12.5 point, which matches what printers have specified for over a century. De Vinne listed 10, 11 and 12 point as the working sizes for octavos and duodecimos.
Leading. Set text solid and it becomes hard work. The 1906 Chicago style manual noted that "nearly all books are leaded", with two points of extra space as the standard. In modern terms, line spacing somewhere around 1.3 to 1.5 times the type size is a reasonable starting point, and the W3C's accessibility guidance asks for at least 1.5 within paragraphs.
Line length. The 1913 report gave a maximum measure of about 93 mm for adult reading. Gress offered a rule of thumb that still works: set a line to about one and a half lowercase alphabets of your chosen face. The W3C's limit is 80 characters.
Print and Screen Are Genuinely Different
Faces designed for screens have systematically larger x-heights relative to their body size. The same 2011 study measured the progression: hardcover books 0.42, paperbacks 0.43, newspapers 0.48, print faces adapted for screens 0.49, and faces designed for screens 0.52.
That is why a face that looks right on a screen can look oversized and coarse on paper, and why a delicate book face can look weak on a screen. If you are producing both a print edition and an ebook, the print interior is where the typeface choice actually sticks. In a reflowable ebook the reader usually overrides it anyway.
Accessibility, Where the Evidence Is Strongest
Low vision. Size and spacing carry almost all the benefit. The 2020 review found that attempts to help by adjusting typeface properties "proved to be futile, with any improvement being modest at best". For accessible editions, UKAAF gives 12 point as the clear-print minimum with 14 recommended, and 16 to 18 point for large print.
Dyslexia. This one is worth stating clearly, because the popular advice is wrong. Multiple independent studies have found that specialist dyslexia typefaces do not help. One found the Dyslexie font "neither benefits nor impedes the reading process of children with and without dyslexia", and that children actually preferred Arial and Times. Another found OpenDyslexic "produced negative results, or decreased students' outcomes compared to both Arial and TNR".
The most useful finding explains why the fonts sometimes appear to work. When researchers matched a standard font's letter and word spacing to the specialist font's spacing, "the difference in reading speed was no longer significant". The benefit was never the letterforms. It was the spacing.
That is actionable: increasing line spacing and letter spacing is the evidence-backed version of switching to a dyslexia font.
One caution on face selection that does have support: avoid faces whose characters are hard to tell apart. The DAISY Consortium warns against fonts "that represent capital I, lower-case L and the number 1 as identical (or near-identical) characters".
Doing This in Deckle
Deckle ships 24 body fonts, including a solid set of serif text faces: EB Garamond, Libre Baskerville, Crimson Pro, Source Serif 4, Spectral, Lora, Cardo, Gentium Book Plus, Ibarra Real Nova and Cormorant Garamond, plus Bodoni Moda for a high-contrast modern look. Any of these is a defensible choice for a novel interior.
Two are worth singling out given the research above. Atkinson Hyperlegible, designed by the Braille Institute specifically to make characters easier to distinguish from one another, targets exactly the failure mode DAISY warns about. OpenDyslexic is also included, and the evidence says its benefit comes from spacing rather than its letterforms, so if you are reaching for it, also increase the line height.
Line height, paragraph spacing, indent and hyphenation are all separate controls, which matters because they are the levers that actually move readability. Given the evidence, adjusting line height is likely to do more for a reader than swapping between two respectable serif faces.
Type size and typeface interact with everything else in the interior: they drive the page count, which drives the gutter and the spine width. Settle the type before you finalise the specification.
Common Mistakes
- Believing serifs are proven more legible. No study isolating serifs has found an advantage, and accessibility bodies say so directly.
- Comparing faces at the same point size. Compare at matched x-height, or the comparison is meaningless.
- Choosing a display face for body text. What looks striking in a line is punishing over 300 pages.
- Setting text solid. Books have been leaded as standard for over a century.
- Reaching for a dyslexia font instead of adjusting spacing. The evidence points at spacing.
- Using a screen-designed face for print. Screen faces carry larger x-heights and can look coarse on paper.
FAQ
Q: What size should body text be? A: Around 10 to 12 point for a novel, which is where measured modern books sit and where printers have specified for a century.
Q: Serif or sans for a novel? A: Serif, by convention and reader expectation. Not because it is measurably more legible, since the evidence does not support that.
Q: Does a dyslexia font help? A: The research says no. What helps is wider letter and line spacing, which is what those fonts happen to build in.
Q: How much line spacing? A: More than none. Around 1.3 to 1.5 times the type size is a sound starting point; accessibility guidance asks for at least 1.5.
Q: Is Garamond a good book face? A: It is a traditional one, though note that most "Garamond" fonts descend from a later punchcutter, Jean Jannon, rather than from Claude Garamont. Daniel Berkeley Updike, writing in 1922, did not rate the version available to him for book text at all.