Kerning adjusts the space between one specific pair of letters, such as "AV" or "To", using values stored in a table inside the font; tracking (letter-spacing) adds or removes the same amount of space between every letter in a run of text; leading is the vertical distance from one baseline to the next. Kerning and tracking are measured in thousandths of an em, so they scale with the type size, and leading is usually set at 120% to 145% of the type size for body text.
- Kerning fixes one letter pair, from a table the type designer stored inside the font.
- Tracking adds the same space to every gap; in design apps it is counted in 1/1000 em.
- Leading of 120% to 145% of the type size suits body text (Butterick).
- Roboto closes the pair "LT" by 0.134 em; lowercase "av" gets only 0.007 to 0.02 em.
- Lato stores 63,819 individual kerning pairs; Noto Sans only 1,436.
- Squeezing letters below default spacing measurably slows early letter recognition (Civera et al., 2024).
Glossary posts stop there. I wanted to see what is really inside a font file, so I downloaded five open-source families from the Google Fonts repository (Inter, Roboto, Lato, Source Serif 4 and Noto Sans), read their tables with the fontTools library and shaped text with HarfBuzz, a widely used open-source shaping engine. Every kerning value in this post comes from those files.
What is the difference between kerning and tracking?
Kerning is a pair-specific correction designed into the font; tracking is a uniform adjustment you apply to a whole word, line or paragraph. Kerning makes "AV" sit as evenly as "HH"; tracking makes all-caps headings breathe or tightens a display line.
Steve Jobs left Reed College in Portland after about six months, then kept sitting in on classes for roughly 18 more months. One of them was calligraphy. In his Stanford commencement address of 12 June 2005 he said the class taught him about serif and sans serif type and about varying the space between different letter combinations, which is kerning by its plain description.
Ten years later, designing the first Macintosh (1984), he said it all came back. The Mac shipped with multiple typefaces and proportionally spaced fonts, and Jobs argued that without that dropped-out year of calligraphy, personal computers might never have had them. Every number in the tables below is the descendant of that idea: spacing as a designed decision, not a fixed slot.
Steve Jobs, Stanford commencement address, 12 June 2005 (full transcript)| Kerning | Tracking (letter-spacing) | Leading (line spacing) | |
|---|---|---|---|
| Direction | Horizontal | Horizontal | Vertical |
| Applies to | One specific glyph pair | Every gap in the selection | Every line in the paragraph |
| Who sets it | The type designer, inside the font | You, in the layout | You, in the layout |
| Unit | Font units (fractions of an em) | 1/1000 em in InDesign and Figma-style tools, em or px in CSS | Points, or a ratio of type size |
| CSS property | font-kerning | letter-spacing | line-height |
| Where it lives | OpenType GPOS table, 'kern' feature | The renderer | The renderer (font supplies default ascent and descent) |
You can also "kern" by hand, called manual kerning, by nudging a single gap in a logo or headline. That is still kerning: one pair, one value. Tracking a word by +50 and then fixing one gap by −30 is the normal way to finish a wordmark.
What is an em, and why do spacing values use it?
An em is the font size itself: in 16 px type, 1 em is 16 px. Fonts are drawn on a grid of "units per em" (UPM), and every width, kerning value and metric is stored in those units, which is why spacing scales automatically when you change size.
The UPM differs between fonts, so raw numbers are not comparable until you divide by it:
- 2,048 units per em: Inter and Roboto;
- 2,000: Lato;
- 1,000: Source Serif 4 and Noto Sans.
A kerning value of −140 in Inter is −140/2,048 = −0.068 em, which at 16 px pulls the second letter 1.1 px closer. Layout tools then express tracking in 1/1000 em, so "−20" tracking is −0.02 em, or −0.32 px at 16 px.
Calculated in this post from Inter's 2,048 units per em, read from the Inter font files in the Google Fonts repository: −140 / 2,048 = −0.068 em, and 0.068 × 16 px ≈ 1.1 px.

Where does kerning live inside a font file?
In the GPOS (glyph positioning) table, as pair-adjustment lookups attached to the 'kern' feature. The OpenType specification defines two formats: Format 1 lists individual glyph pairs, and Format 2 groups glyphs into classes so one value covers, say, every "T" followed by any round lowercase letter.
Each stored value is an XAdvance in font units: negative pulls the next glyph closer, positive pushes it away. Class kerning keeps files small; individual pairs handle exceptions. Here is what the five fonts contain, counted from the default instance of each file.
| Font (Google Fonts repo, 2026) | Units per em | Glyphs | Individual kerning pairs | Non-zero class pairs |
|---|---|---|---|---|
| Lato Regular | 2,000 | 3,023 | 63,819 | 1,985 |
| Inter (variable) | 2,048 | 2,933 | 13,472 | 1,290 |
| Roboto (variable) | 2,048 | 1,326 | 7,450 | 358 |
| Source Serif 4 (variable) | 1,000 | 1,463 | 2,300 | 5,321 |
| Noto Sans (variable) | 1,000 | 4,515 | 1,436 | 1,241 |
These counts are not a quality score. Source Serif 4 has few individual pairs but over 5,000 class pairs, and each class pair can stand for dozens of glyph combinations. Lato simply stores most of its decisions as explicit pairs. What the table does show is the scale: a professional text face carries thousands of spacing decisions you never see.
What are the most common kerning pairs?
Pairs where a diagonal or overhanging letter meets an open one: AV, VA, AW, To, Ty, Yo, LT, P., T, and similar. The answers on the Graphic Design Stack Exchange thread about common pairs list the same families of shapes. Here is how much four fonts actually kern them, converted to thousandths of an em so the fonts can be compared.
See the data as a table
| Inter (/1000 em) | Roboto (/1000 em) | Lato (/1000 em) | Source Serif 4 (/1000 em) | |
|---|---|---|---|---|
| AV | -68.4 | -42.5 | -54 | -119 |
| To | -78.1 | -48.3 | -122 | -70 |
| Ty | -62.5 | -35.2 | -86.5 | -40 |
| LT | -96.2 | -134.3 | -94 | -80 |
| Yo | -76.7 | -31.7 | -109.5 | -90 |
| av | -14.6 | -7.3 | -10.5 | -20 |
Source: My measurement: HarfBuzz shaping with and without the kern feature on font files from the Google Fonts GitHub repository (google/fonts), Nov 2026 · Negative values pull the second letter closer. Variable fonts measured at their default instance.
Two patterns stand out:
- Capital pairs are kerned hard. Pairs like LT can take more than a tenth of an em, enough to be obvious at headline size; Roboto closes LT by 0.134 em.
- Lowercase pairs barely move. Pairs like "av" get only 0.007 to 0.02 em, because lowercase letters are spaced well by their sidebearings already.
If your software has kerning switched off (some email clients and older apps do), the "To" in a heading opens up by up to 0.12 em, which is why it looks gappy.
Metrics or optical kerning?
There are two kinds of automatic kerning in design apps. "Metrics" kerning uses the pairs in the font, as measured above. "Optical" kerning ignores the table and computes spacing from the outlines; it helps for fonts with poor tables or for mixing fonts, but it overrides a good designer's decisions, so use metrics for well-made text faces.
| Mode | How it spaces | Use it for |
|---|---|---|
| Metrics | The pairs stored in the font | Well-made text faces (the default choice) |
| Optical | Computed from the outlines; ignores the table | Fonts with poor tables, or mixed fonts in one line |
Why do round letters look wrong when aligned exactly?
Because the eye judges size by area, not by the extreme points. A circle that touches the same lines as a square looks smaller, so type designers make round letters overshoot the baseline and cap height slightly. Measured from the files, the capital O in Inter extends 20 units (1.3% of cap height) below the baseline and 20 units above the cap height.
| Font | Cap height (units) | O below baseline | O above cap height | o below baseline |
|---|---|---|---|---|
| Inter | 1,490 | 20 (1.3%) | 20 (1.3%) | 24 |
| Roboto | 1,456 | 20 (1.4%) | 20 (1.4%) | 20 |
| Lato | 1,433 | 15 (1.0%) | 16 (1.1%) | 14 |
| Source Serif 4 | 670 | 15 (2.2%) | 15 (2.2%) | 13 |
This is the subject of the Stack Exchange question "Aligning letters wrong appears more right", and the same principle drives spacing. A round letter next to a straight one needs less side space than two straights, because the curve already leaves white space at the top and bottom. Sidebearings (the space built into each side of a glyph) encode that, and kerning corrects the cases sidebearings cannot.
When should you change tracking?
Track out all-caps and small sizes, track in large display sizes, and leave body text alone:
- All caps and small labels: track out (see the questions below for typical values);
- Large display headlines: track in slightly;
- Body text: leave it at the font's default.
Fonts are spaced for a typical reading size; capitals were designed to sit next to lowercase, so a line of capitals looks cramped, and big headlines look loose because the same proportional gaps become physically wide.
What the reading research says
The reading research backs the "leave body text alone" part. Extra letter spacing makes individual words easier to pick out but does not make adults read faster, and squeezing letters below the default spacing does measurable harm.
An eye-tracking study of 24 German university students reading texts set in 12-point Times New Roman, either at normal spacing or with letter spacing increased by 2.5 points. Wider spacing shortened first-fixation and gaze durations on each word, most of all for slower readers. But readers skipped fewer words and made more fixations, so words read per minute did not improve, and the faster readers were actually slowed down. Crowding sensitivity did not predict who benefited. The limit: one font, one spacing increase, and a small sample of skilled adult readers.
Read the paper (free)Thirty-nine students at the University of Valencia (eight were excluded from analysis) read Spanish words in 36-point Times New Roman at default spacing, condensed by 1.5 points, or expanded by 1.5 points, while EEG was recorded during a semantic categorisation task. The N170, a brain response tied to early letter encoding, was largest for condensed words; default and expanded spacing did not differ. The authors' reading: tightening letters below the default makes early letter identification harder, while loosening them gives no extra benefit. The limit: isolated words at a large size, not running text.
Read the paper (free)A third study, by Galliussi and colleagues in Annals of Dyslexia (2020), tested 64 children with and 64 without dyslexia and found that adding letter spacing without also adding word spacing slowed reading. The practical rule follows: if you track text out, open the word space too, or words start to blur into each other.
How much leading and line length should body text have?
Set leading at 120% to 145% of the type size and keep lines between 45 and 90 characters, including spaces. Those are the ranges Matthew Butterick gives in Practical Typography, and they work together: longer lines need more leading so the eye can find the start of the next line.
Where the word and the numbers come from
The word comes from the strips of lead that compositors slid between lines of metal type, and the notation survives: "10/12" means 10-point type on 12-point baselines, so 2 points of lead.
In CSS, line-height sets the baseline distance and the browser splits the extra space as "half-leading" above and below each line. Word's "single" spacing is about 117% (Butterick), which is slightly tight for body text by his 120% to 145% range.

Why x-height changes the leading you need
Fonts with a large x-height (lowercase height relative to the em) look bigger and need more leading at the same point size. From the files: Inter's x-height is 1,118/2,048 = 0.546 em, Noto Sans 0.536 em, Lato 0.507 em, Source Serif 4 only 0.475 em.
Calculated in this post from x-height and units-per-em values read from the font files in the Google Fonts repository, for example 1,118 / 2,048 = 0.546 em for Inter.
| Font | x-height (em) |
|---|---|
| Inter | 0.546 |
| Noto Sans | 0.536 |
| Lato | 0.507 |
| Source Serif 4 | 0.475 |
Inter at 16 px therefore has lowercase letters about 15% taller than Source Serif 4 at 16 px, which is why swapping fonts without adjusting size and leading changes the whole page. CSS font-size-adjust, Baseline since July 2024, can normalise x-height across fallback fonts.
Use this to check what you just read: plug in your own column width and type size.
Characters per line and line height
Line length and leading are also the backbone of any layout grid; I build both into a column system in the typographic grid system guide. Whether serif or sans actually reads better at these settings is a separate argument, covered in serif vs sans serif readability.
Does kerning work the same way in Hindi and other Indian scripts?
Partly. Devanagari fonts also have a 'kern' feature, but much of their spacing is decided earlier, when the shaping engine builds conjuncts and positions matras. Noto Sans Devanagari carries 486 individual kerning pairs and 322 non-zero class pairs, far fewer than Latin fonts, because the shirorekha (headline) joins letters and leaves less room for pair-by-pair fixes. Tracking Devanagari breaks that headline into pieces, so avoid letter-spacing Hindi text. The full mechanism is in how Devanagari fonts work.

- Set it large. Spacing errors hide at small sizes.
- Settle the tracking first. Get the density of the whole word right.
- Flip or blur it. Upside down or squinting, you see shapes, not words, and odd gaps jump out.
- Kern only those pairs. Fix them one at a time with manual kerning.
Squeezing letters below the default spacing does measurable harm; loosening them gives skilled readers no extra speed.Summary of Korinth et al. (2020) and Civera et al. (2024)
I never went to design school; my design work has grown alongside freelancing since about 2013. More on that path and the rest of the studio at Kala.
What's new
- Google open-sources Google Sans Flex, a variable font with weight, width, optical size, slant, grade and roundedness axes Google Design, 2025
- CSS text-box trimming, which removes the extra space above and below text so leading can be set precisely, ships in Chrome 133 and Safari 18.2 web-features, Feb 2025
- CSS font-size-adjust, for matching x-heights across fallback fonts, is on track to become widely available in January 2027 web-features, 2026
Questions people ask
Is kerning the same as letter-spacing?
No. Letter-spacing (tracking) changes every gap by the same amount. Kerning changes only specific pairs by amounts the type designer chose, so "AV" closes up while "HH" stays put. In CSS, letter-spacing is tracking and font-kerning switches the font's kerning on or off.
Should kerning be on for body text on the web?
Yes. Browsers apply the font's kerning by default for most text, and it costs almost nothing. Turning it off makes capital pairs and punctuation gaps uneven, with differences of up to about 0.13 em in the fonts I measured.
What is a good line height for 16 px text?
About 19 to 23 px, which is 1.2 to 1.45 times the size. Use the higher end for long lines, fonts with a large x-height, and screens; the lower end for short lines and headings.
Calculated in this post from Butterick's 120% to 145% range applied to 16 px: 16 × 1.2 = 19.2 px and 16 × 1.45 = 23.2 px.
Why does all-caps text need extra tracking?
Capitals are spaced to sit next to lowercase letters, which have open shapes and ascenders that leave white space. A run of capitals loses that space and looks dense, so a small amount of positive tracking, often +50 to +100 thousandths of an em, restores even colour. This is a convention among designers rather than a measured standard, so judge it by eye.
Sources
- Microsoft Typography: OpenType specification, GPOS table (Pair Adjustment Format 1 and 2, design units)
- Google Fonts GitHub repository (font files for Inter, Roboto, Lato, Source Serif 4, Noto Sans; my measurements with fontTools and HarfBuzz, Nov 2026)
- HarfBuzz documentation
- Butterick's Practical Typography: Line length (45 to 90 characters)
- Butterick's Practical Typography: Line spacing (120% to 145%)
- Wikipedia: Leading (origin, 10/12 notation, CSS half-leading)
- Korinth, Gerstenberger and Fiebach: Wider letter-spacing facilitates word processing but impairs reading rates of fast readers, Frontiers in Psychology (2020)
- Civera et al.: The effect of inter-letter spacing on the N170 during visual word recognition, Cognitive, Affective, & Behavioral Neuroscience (2024)
- Steve Jobs: 2005 Stanford commencement address, transcript (the calligraphy story)
- Galliussi et al.: Inter-letter spacing, inter-word spacing, and font with dyslexia-friendly features, Annals of Dyslexia (2020)
- Graphic Design Stack Exchange: Difference between kerning vs letter spacing?
- Graphic Design Stack Exchange: What are some common kerning pairs?
- Graphic Design Stack Exchange: Aligning letters "wrong" appears more "right"
- Google Design: Making Google Sans Flex (2025)
- web-features explorer: text-box and font-size-adjust
