Mastering the Art of Measuring and Testing Kerning Accuracy

Kerning accuracy can make or break a typeface’s personality. A single misplaced gap can turn elegant text into a visual stumble.

Designers who master kerning measurement don’t just “eyeball” spacing—they use repeatable systems, calibrated tools, and live tests across devices. The result is typography that feels invisible because it simply works.

Understanding Kerning vs. Tracking vs. Letter-fit

Kerning adjusts the white space between two specific glyphs, while tracking alters spacing across an entire string. Letter-fit is the macro decision that determines how a typeface was spaced before any manual kerning began.

A common rookie error is to tighten tracking on a headline and assume the job is done. That global move ignores the unique collisions that only occur between particular pairs like “Te” or “Av.”

Imagine a luxury logo set in Bodoni. If you track the whole word inward, the ball terminals of the “y” and the sharp apex of the “A” still crash. Only pair-based kerning can rescue that junction.

Human Perception and the Illusion of Equal Space

Our eyes judge balance by color density, not by ruler distance. A round “O” next to a straight “H” can measure 50 units on each side yet feel lopsided because the curves reflect more light.

Designers compensate by giving curved glyphs slightly less numerical space. The perceived gap looks identical even though the actual gap is narrower.

Testing with the Blur Trick

squint at the text or apply a 3-pixel Gaussian blur in Photoshop. If any pair suddenly darkens or lightens, the kerning is off.

This low-tech hack forces you to see tonal weight instead of letterforms. It is fast, works at any zoom level, and reveals errors that a 400% artboard view can hide.

Establishing a Kerning Reference System

Before touching a single pair, create a “master string” that contains every combination you will ever need. A classic Latin set is “nnnonnonoHaxoHaxoNoNo” but you can extend it to include diacritics, punctuation, and numerals.

Store that string as a GlyphSet in Glyphs App or as a Character Set in FontLab. One click repopulates any new master so you never miss a pair.

Next, assign reference glyphs. The “n” and “o” are traditional because their shapes represent straight, round, and diagonal archetypes. Lock their side-bearings first; every other glyph is then spaced relative to them.

Digital Tools That Measure Kerning Objectively

MetricsMachine streams live numeric values as you drag a glyph left or right. The readout updates to 1/1000 em precision, removing guesswork.

FontLab’s Kerning Conflict panel flags classes that overwrite each other. Resolve these early or late-stage interpolations will explode.

RoboFont’s KernSlider displays all pairs in a heat-map: green for loose, red for tight. You can spot systemic errors across entire families in seconds.

Scripting Micro-Adjustments

Python scripts can batch-adjust kerning by percentage. Tighten every pair by 2% for the caption master while leaving display untouched. This keeps optical size logic intact without manual repetition.

Store the delta as a custom parameter so future exports inherit the tweak automatically.

Optical Size and Kerning Delta

A 6 pt paragraph needs looser kerning than a 72 pt headline because ink spread closes white space at small sizes. Type designers build “size-specific kerning” into variable fonts using an opsz axis.

Test this by exporting a single variable file and moving the slider in Adobe Illustrator. Watch the kerning values shift in real time—proof that the font carries discrete spacing data for each optical size.

If your toolchain does not support opsz kerning, maintain separate .ufo masters and merge them later. Label masters clearly to avoid pushing caption spacing into a billboard.

Cross-Platform Rendering Tests

Windows GDI ClearType ignores fractional kerning values below 0.5 font units. A pair set to –25 on 1000 UPM renders as –24 or –26 depending on the driver version.

Export a dummy web page that loads the font via WOFF2 and view it in Edge, Chrome, and Firefox on both high-DPI and 96 ppi monitors. Screenshot each browser at 14 px and overlay the captures in difference mode. Any misalignment larger than one pixel signals a rounding error you must compensate for at the source.

iOS CoreText rounds kerning to integer units as well, but the threshold is 0.25. Test on an iPhone SE first; its lower resolution exaggerates rounding artifacts.

Variable Fonts and Kerning Compression

Traditional pair tables balloon file size when you support 3000 pairs across four axes. Use glyph-level variable kerning: store only deltas from a default axis position.

For example, if the “T–a” pair is –80 at wght 400 and –120 at wght 700, encode –40 as the delta. This cuts table size by 60% and speeds browser parsing.

Validate with Google Fonts’ fontdiffenator. It highlights kerning drift between any two versions and exports PNG diff strips you can drop into a GitHub issue.

Proofing with Real Words, Not Strings

“Hamburgefontsiv” is a decent stress test, but it omits rare pairs like “f–?” or “–72”. Generate a proof that scrapes your client’s actual product names, cities, and surnames.

A skincare brand once discovered its star serum “Phytoxyl™” looked broken on shelf packaging because the “y–l” pair was never kerned. A real-word proof caught the flaw before 100 k labels were printed.

Use OpenType feature code to inject discretionary ligatures or alternates that appear only in those real words. This keeps the default set clean while solving bespoke collisions.

Automated Regression Testing

Set up a CI pipeline that renders every commit into 200 PNGs of critical pairs. Pixel-match them against the previous build using ImageMagick’s compare metric.

A drift score above 0.02% fails the build and posts the diff to Slack. This prevents accidental kerning regressions when you rename glyph classes or merge branches.

Store the reference PNGs as LFS objects so designers can pull them locally and inspect failures at 8× zoom without running the full script.

Hand-Kerning Versus Class-Kerning Strategy

Class-based kerning groups “T” with “Ť” and “Τ” (Greek) so one value covers dozens of pairs. Over-grouping, however, creates new problems: the Cyrillic “Э” may need more space than the Latin “E”.

Start broad: create 30 classes for straight-left, round-right, etc. Then hand-kern the top 2000 pairs and let analytics reveal which class exceptions save the most data.

FontTools’ pyftsubset can output a kerning report sorted by frequency. Promote the top 5% of pairs to hand-kerned status and leave the rest classed. You gain both precision and a smaller binary.

Kerning for Accessibility and Legibility

Readers with dyslexia rely on consistent glyph rhythm. Sudden tight pairs like “rn” morphing into “m” can hijack word recognition.

Test in a dyslexia simulator such as Lexend and maintain a minimum 40-unit gap between potentially ambiguous sequences. This may feel loose to a graphic designer, but it measurably reduces fixation time for dyslexic users.

WCAG 2.2 does not specify kerning values, but it mandates adequate letter spacing for captions. Proactively supply a user stylesheet that overrides kerning with font-kerning: none and adds 0.05 em letter-spacing. This empowers low-vision users without ruining your default aesthetic.

Print vs. Digital Kerning Calibration

Ink gain on uncoated stock closes gaps by 3–5%. A pair that looks crisp on Retina may clog on newsprint.

Print a 600 dpi laser proof and examine it under a 10× loupe. If counters start to fill, loosen the relevant pairs by 5 units and reprint. Iterate until the loupe view matches the screen at 100%.

Conversely, glossy stock reflects more light and makes spacing appear looser. Tighten pairs by 2–3 units for high-gain coatings like UV varnish to maintain the same perceived density.

Legal and Brand Risks of Poor Kerning

A pharmaceutical label that mis-kerningly fuses “1 mg” into “Img” can trigger FDA rejection. The recall cost dwarfs the price of proper type QA.

Trademark offices require wordmarks to match the registration specimen exactly. If kerning drifts in a rebrand, you may forfeit protection.

Build a kerning lockfile: export a signed .plist that records every pair value. Embed its hash in the font name table so future audits can prove the spacing is unchanged.

Workflow Checklist for Release-Ready Kerning

Run this list before every export: 1) MetricsMachine audit for overlapping classes, 2) Browser diff at 13, 16, 21 px, 3) Loupe print on specified stock, 4) Dyslexia simulator pass, 5) CI pixel regression green, 6) Client real-word proof signed off.

If any step fails, bump the minor version and retest. A font that passes all six gates is statistically unlikely to embarrass you in the wild.

Archive the checklist log alongside the .otf. Two years later, when the brand expands to Cyrillic, you will know exactly which kerning standard to match.