How to pair fonts
A short method. Every chapter points at live pairings in the library, so the argument is always something you can look at.
First draft, 27 September 2026. The chapters will grow as the library does.
- 1. Why two fonts
- 2. The four ways a pair holds together
- 3. Match the proportions
- 4. Assign the roles
- 5. Let scale do the work
- 6. Print and screen
- 7. The usual failures
- 8. How to read the examples
- 9. Science and journals
- 10. Whimsy and wonder
- 11. Comics and graphic novels
1. Why two fonts
One font can do a whole job. Weight and size alone give you headline, subhead and text, and some of the best-set books use one family throughout. A second font is a decision, not a default, and it is made for one of two reasons. Either the job has two voices, a shouting one and a reading one, and one family cannot stretch that far. Or the job wants a tone the reading face cannot supply, so a display face is brought in for the headlines only.
A third font is almost always a mistake. The exceptions are a monospace for code and data, and a script used as a small accent. Both appear in the library as an accent, never as a third equal.
2. The four ways a pair holds together
Every pairing that works can be explained by one of four principles. The library tags each pairing with the one it uses.
Contrast. Two different classes: a serif over a sans, a slab over a grotesque, a script over a geometric. The difference is obvious at a glance, so the reader never confuses the two roles. This is the most common principle and the easiest to get right, provided the two faces agree on proportions (chapter 3).
Concord. Two faces from the same class that agree in spirit: two humanist sans, two old-style serifs. Quieter than contrast and harder to pull off, because the reader has to sense a difference that is not obvious. It works when the two differ clearly in weight or width.
Superfamily. Two faces drawn together on one skeleton: PT Serif and PT Sans, IBM Plex, Source, Merriweather and its sans. The pairing is guaranteed by construction. Choose this when the type has to be right first time and nobody has time to fuss.
Single family. One family, with weight and size doing all the work. Not a pairing at all, and included here because it is often the correct answer.
3. Match the proportions
Contrast in class, agreement in proportion. The two things to check are x-height and overall width. Set the same sentence in both faces at the same size and look at the lowercase: if one x-height is visibly taller, the pair will fight every time they sit on the same line. Then look at width: a wide geometric sans over a narrow old-style serif looks like two separate documents pasted together.
The exception is when the difference is the point. A very wide display face over a very narrow text face reads as deliberate when the size gap is large enough (chapter 5).
4. Assign the roles
The display face gets the short, large text: titles, section names, one-line claims. The text face gets everything that is read for longer than a few seconds. Decide this first and the rest follows. A face that is beautiful at 72 pixels and unreadable at 14 is not a bad face; it is a display face, and it should never be given the body.
Cormorant Garamond is the standard example. Its x-height is tiny, so it disappears at text size, but at 40 pixels it is one of the most atmospheric faces on Google Fonts. Every Cormorant pairing in the library gives it the headline and gives the reading to something sturdier.
5. Let scale do the work
A pairing with a small size gap has to rely on class contrast to separate the roles. A pairing with a large size gap can get away with almost anything, because a 96 pixel headline and a 15 pixel paragraph are never confused. The café and bar menus from the trending pairings are all built this way: a massive section title in a wide or wonky face, then a quiet small face for every item. The gap in scale is the design.
Weight is the second lever. A heavy headline over a regular body is the safe setting. A light headline over a regular body needs a display face with enough presence to hold the top of the page at low weight, usually a Didone or a wide sans.
6. Print and screen
On screen, tall x-heights and open apertures win at small sizes, which is why Inter, Open Sans and Lato are the default body faces. In print, a serif with moderate contrast reads more comfortably across a wide measure, and the high-contrast Didones that look fragile on a phone look expensive on heavy stock. Textured and recycled paper flattens fine detail, so wide open faces survive it and hairlines do not.
The library tags each pairing as screen, print or both, and the print pairings say which paper they expect.
7. The usual failures
- Too similar. Two neutral sans of the same width and x-height. The reader senses a change but cannot see it, which reads as a mistake.
- Two display faces. Both want attention, neither can be read for long, and the page has no rest.
- Display face in the body. Cormorant at 14 pixels, Playfair for paragraphs. Always a display face's fault-line.
- Three equal fonts. If there is a third, it is an accent with one job: code, prices, a signature.
- Same weight throughout. With no weight gap the roles blur even when the classes differ.
8. How to read the examples
Each pairing page opens with the two faces set together, then gives a styling recipe in real numbers, the reason it holds together, and seven layouts: menu, article, app screen, poster, stationery, report and landing page. The layout the pairing was made for comes first. Look at that one, then look at the article and the app, because those two show whether the body face can actually read at length and at small size.
Pairings come from published guides, credited on each page, and from our own library. The recipe on an imported pairing is generated from its weights until someone writes it by hand, and the page says so.
9. Science and journals
Scientific text is the hardest case for a pairing, because the body has to carry Greek letters, operators, subscripts and author names in three scripts, for an hour at a time, and the figures have to stay legible at eight points. The rules that follow are what the journals themselves do.
Serif for the paragraphs, sans for the figures. Nearly every journal sets the reading in a serif and every axis label, legend and callout in a sans. PLOS ONE permits only Times, Arial and Symbol; Cell Press asks for Avenir or Arial inside figures. Nature commissioned its own serif, Harding, for the 2019 redesign. Schwen's ten rules for scientific typography (PLOS Computational Biology, 2020) say the same: serif for body text, sans for slides and posters, monospace for code, and a line of 75 to 80 characters.
Check the glyphs before the look. A body face for science needs Greek and the mathematical operators drawn in its own style, or formulas will fall back to a system font and the page will show it. On Google Fonts the safe body serifs are STIX Two Text, Noto Serif, Libertinus Serif, Gentium Plus, EB Garamond, Literata, Piazzolla, Alegreya and Source Serif 4. STIX Two Math, Noto Sans Math and Libertinus Math are the dedicated formula faces. Charis SIL carries the full phonetic alphabet.
Superfamilies earn their keep here. A table that mixes a serif caption, a sans header and a mono column of numbers reads as one document only when the three share metrics. IBM Plex, Source, Noto, Libertinus, Merriweather and Alegreya all ship a matched set, and several are what a LaTeX user would reach for once Computer Modern feels too thin.
Mono for the data. Fixed-width digits line up in columns without tabular-figure tricks, and identifiers, accession numbers and code should never be set proportionally. IBM Plex Mono, Roboto Mono, JetBrains Mono and Libertinus Mono are the ones in the library.
Beauty is allowed. The journal that wants to look like a magazine uses a Didone or an inscriptional face for titles and keeps the rules above for everything else. Cinzel over Spectral, Playfair over Roboto, Syne over Cormorant: the display face is a masthead decision, not a reading one. Lean on the italics of Cormorant and Lora for abstracts and species names, and track a bold or classical heading face a little wider than feels natural.
10. Whimsy and wonder
Fairy tales, grimoires and Wonderland ask the opposite of a journal: the display face is allowed to be a costume. The discipline moves to the body, which has to stay plain enough to be read aloud at bedtime, and to the rule that a costume face gets one job and a size floor.
Pick the era, then the face. Google Fonts has a real range here: the Fell types cast in the 1680s with the ink spread left in; uncial and blackletter for the illuminated and the Germanic; Victorian storybook serifs like Alice, which was drawn with Carroll's book in mind; art nouveau labels; copperplate and chancery scripts; and circus faces that wobble on purpose. Each says a specific century. Choose the century the story lives in, and the face chooses itself.
The body carries the tale. Under any of these, an old-style book serif does the reading: EB Garamond, Crimson Pro, Cardo, Alegreya, Alice itself. Cormorant Garamond can be the body only at Medium weight and 13 pixels or more. A rounded sans works under the sugary and circus faces, where a serif would be too grown up.
Concord is common here. Many of these pairs are serif over serif: a carved or scripted display face over a printed book face. The contrast comes from era and construction rather than class, and it works because the display face is unmistakably not the text face.
Track the capitals, lose the black. Inscriptional and gothic heading faces want their letters given room, 0.1 to 0.16em. And nothing in this chapter should be set black on white: charcoal on parchment, cream on plum, gold on emerald. Two of the library's palettes exist for exactly that.
11. Comics and graphic novels
A comics page has more typographic roles than any other kind of page: the logo, the story title, the balloons, the caption boxes, the sound effects, the credits. The pairing model still holds if you assign them properly. In this library's comics pairings, the heading is the cover or logo face, the body is the balloon lettering, and an accent, where there is one, is the sound effect or the caption.
The balloon face is the real decision. It is read for the whole book, so it has to be a lettering hand that stays even at 13 pixels: Comic Neue, Patrick Hand, Mali, Short Stack, Itim, Kalam. Comic Neue is Comic Sans redrawn with its rhythm fixed, and it is the right default; the stigma belongs to the original. Caps or mixed case is a house choice. American mainstream letters in caps; most of Europe and nearly all webcomics use mixed case.
The logo face gets one job. Bangers, Bungee Shade, Bowlby One, Creepster, Metal Mania, Rye, Shojumaru: each of these is a whole genre in one weight, and each belongs on the cover and the chapter card only. Never in a balloon.
Captions want a third voice. Narration boxes read best in a face that is clearly not the balloons: a typewriter mono for crime and horror, a condensed grotesque for pulp and war, a book serif for a graphic memoir. That is where the serif-over-sans thinking from the rest of this guide comes back in.
Manga is its own system. Heavy gothics for shōnen, rounded pop hands for shōjo, minchō serifs for gekiga. Google Fonts has full Japanese coverage in Dela Gothic One, Zen Kaku Gothic New, Hachi Maru Pop, Kaisei Decol and Zen Old Mincho, so one pair can set the original and the translation.
For the drawing styles these pairings sit beside, the Comic Art Styles field guide on the books shelf covers forty-three of them.