Design Fundamentals: Kerning and Tracking

Adapted from The Complete Manual of Typography: A Guide to Setting Perfect Type, 2nd Edition (Adobe Press)

By Jim Felici

In setting type, it’s often the little things that count. Two of the smallest adjustments you can make are those that move individual characters closer together or farther apart. They’re also two of the most important. Although both affect the same thing—the amount of space that characters are allotted on the page—kerning works only on specific letter pairs, while tracking works on ranges of characters, even whole documents.

Definitions: Kerning and Tracking

Kerning and tracking adjustments are calibrated in the same units—usually thousandths of an em—but their goals are quite different.

Kerning is targeted: It adjusts the spaces between specifc letter pairs to adjust anomalies in spacing created by the shapes of the two letters, which can make the characters look too far apart or too close together. The term comes from the archaic French word carne, which means “corner” (these days, it’s a slang pejorative that could be translated genteelly as meathead). In hand composition with metal type, kerns refer to parts of certain letters that overhang the blocks on which they’re cast. When such characters are digitized, kerns extend beyond the limits of a character’s bounding box (see below).

The fraction bar is a kerning character. On the left is a view from within a font-editing program. The bar’s bounding box is far narrower than the character itself, and the parts of the character that extend beyond the box are kerns. The bounding boxes of adjoining characters abut that of the fraction bar, so that the fraction bar overlaps them. The fraction on the right has been set without any adjustment to the spaces between characters—the overlap is the effect of the kerning nature of the fraction bar.

Tracking is generalized: It uniformly affects the spacing between all the characters in a range of text. Tracking is normally adjusted to compensate for spacing problems caused by changes in point size (especially in very small and very large type). It is also used to adjust badly spaced passages of text.

The term tracking has its roots in one of the original phototypesetting machines, the Photon, which created images of characters by flashing a strobe light through a font that consisted of a rotating film-negative disk. The image was focused through a lens and directed onto photographic film by a prism. The prism was mounted on a toothed track, and it advanced according to the width of the character being imaged. The amount of space allotted to each character, then, was a function of how the prism tracked. These days the track is gone, but the term tracking remains.

Kerning in Practice

The illustration below shows some examples of letter pairs in need of kerning and how they look after they’ve been kerned. The idea of kerning is to create between the kerned characters the same sense of spacing as that prevailing in the surrounding text. The idea is not simply to squeeze out as much space as possible.

The letter pairs that most commonly need kerning are combinations of upper- and lower-case characters and pairs containing punctuation. The To and w. here need kerning because of the open spaces to the frst characters’ right. In the bottom pair, the hook of the f (often a kerning feature) would collide with the quotation marks if the two were not kerned apart.

Most kerning is done automatically in text composed by a page layout or typesetting program. The adjustments in spacing are based on kerning tables built into the fonts. These tables list specific letter pairs and the spacing adjustments—expressed in fractions of an em—that should be applied to them. Usually, these kerning values are negative numbers, meaning that the characters in question should have the space between them reduced. But in some cases, they’re positive values, intended to push certain pairs of characters farther apart.

The number of letter pairs that can be created from a typeface containing “only” 228 characters would run into the tens of thousands (51,984, to be exact). Clearly a comprehensive kerning table is impractical, but it’s also unnecessary, as most of the possible character combinations rarely or never occur, such as a lowercase letter followed by a capital (Irish, Scottish, and software program names excepted).

In general, high-quality fonts contain between 500 and 1,000 kern pairs (see below). Ideally, the values assigned to these pairs have been defined by the typeface designers themselves. There are, however, programs that can create lists of kerning adjustments automatically, and not surprisingly, their results are less than optimal. Reputable font vendors can generally be counted on to have made a serious effort to provide high-quality kerning information.

Because kerning adjustments are expressed in relative units, a kerning adjustment made at one point size will have the same effect when the type is enlarged or reduced.

This list of typical kerning pairs, drawn from a font’s kerning table, consists of more than 600 character combinations. The great majority of them need to be drawn closer together. Note that few pairs comprise two lowercase characters.

Manual Kerning

No matter how extensive a font’s kerning table may be, unlisted pairs that need kerning adjustments will occur. Someone setting type for the television station kqed, for example, will have to reckon with the unorthodox combination Kq. Pairs such as this have to be kerned manually.

Except in unusual circumstances you don’t have to worry about manual kerning in body type. Unless the spacing problem is very great (usually because it’s in an obscure letter combination) less-than-ideal kerning at common text sizes is hardly noticeable.

At display sizes, though, kerning problems become progressively more obvious as the point size grows. This is because, as type grows, white spaces seem to grow faster than the characters around them. In short, irregular spacing becomes more obvious, and even subtle variations become plainly visible. Even kerning pairs that are listed and compensated for within a font’s kerning table may need some hand-kerning in large sizes.

At display sizes, any kerning problems become painfully obvious. In the sample on the left, based on the font’s built-in kerning information, the ru combination is terribly loose, but it can only be tightened so much. The other pairs then have to be adjusted to match its spacing. The values of the adjustments, in thousandths of an em, are recorded below the hand-kerned version on the right.

Page layout programs have keyboard shortcuts for making manual kerning adjustments, and these are usually available in either coarse (say, ¹/20 em) or fine (¹/100 em) increments. There’s rarely any point in bothering with kerning adjustments of less than ¹/100 em. An adjustment of ¹/1000 em is so small that when reproduced on an imagesetter it is either (a) too small to see, or (b) too small to make any difference if it is big enough to see.

In the lower sample here, one pair of characters has been kerned ¹/1000 em closer than in the line above it. Can you tell which one it is? Kerning adjustments this small make such a negligible—if even visible—difference that they’re not worth bothering with.

Manual Kerning Strategies
The first rule of manual kerning is not to do anything until the text has under-gone its final tracking adjustments. All spacing adjustments you make are cumulative, so if you kern your type first and then change its tracking, you’ll be throwing all of your careful adjustments out of whack.

To see type well enough onscreen to kern it effectively, you need to zoom in. The closer you zoom in, the clearer the vision you have of the type, and the more accurate the representation of the spacing will be. The smaller the pixels (relative to the size of the characters), the more closely they resemble the resolution of printed characters—although you’d have to zoom to well over 1,000 percent to approximate even modest imagesetter resolution.

But you can zoom in too close. Remember that as you zoom in—and the size of the type grows—the very spaces you’re trying to adjust are growing deceptively large, exaggerated by their magnifed size. Second, if the idea of kerning is to make spacing appear even over a range of text, you have to have a view of that range to know when you’ve achieved your goal. If you zoom in so close that all you can see are the two letters you’re kerning, you’ll have no frame of reference for your adjustments.

The element that keys all of your hand-kerning work is the toughest, most intractable pair of letters in the whole text. This pair has to set the overall feel for the rest of the spacing. If you make all the other character pairs tighter, the loose one will stick out, as illustrated below.

The software did a fairly good job of kerning the top sample, though the spacing is clearly uneven in spots. In the correctly kerned version in the middle, the overall spacing has been determined by the vi pair, which by its nature creates a somewhat loose spacing feel. In the overkerned bottom sample, the typesetter has squeezed as much space out of every letter pair as possible, but this leaves the vi gap as an obvious hole in the line. The goal of hand-kerning is to create even spacing, not to eliminate spaces.

Having figured out the key pair, you can go about kerning the others. If it appears that nearly all of the pairs need adjustment, do a preliminary adjustment by tightening the tracking of the whole selection. When the key pair is spaced as well as it can be, stop the tracking adjustments and start kerning.

You can’t kern by numbers. You can only kern by what looks right. In other words, measuring won’t help; trust your eyes. If necessary, step back from the screen for a different view, where you can’t see every pixel. Always verify your kerning efforts with printed proofs.

Don’t panic if you reduce the view back down to 100 percent or 150 percent and your kerning work looks dreadful. It’s just a reminder of how limited the concept of wysiwyg really is.

Kerning Italic-roman Character Combinations
You can generally count on a font’s kerning table to accommodate difficult kerning pairs. But when adjoining characters are generated from two different fonts, kerning problems may arise that your program can’t automatically fx.

The most common of these occur when italic characters are set inside roman parentheses or brackets, or when roman apostrophes or quotation marks follow italic words. In these cases, the oblique stance of an ascending italic character can cause it to collide with a character designed for use with upright types. This may be hard to see well onscreen, but you should keep an eye out for situations in which this may occur and zoom in for a closer look.

Transition points from italic to roman types may need kerning, as these examples show. Wherever an ascending italic character is followed by a tall or top-aligning roman character, such collisions are apt to occur.

Algorithmic Kerning

Some programs can kern type not by consulting kern tables inside fonts, but by analyzing the shapes of the character outlines. This algorithmic, or optical, kerning has the advantage of being able to handle any character combination, no matter how uncommon. Its disadvantage is that it doesn’t proft from the experience of the human eye, always the ultimate judge of success.

Compared to the unkerned sample at top, both the table-based and algorithmic kerning routines did a pretty good job. An exception, though, is the algorithmic kerning of the i. If you look at the spaces between the algorithmically kerned i and the characters around it, they measure very close to even. But they don’t look even because the i itself is so narrow. This kind of visual nuance makes human, visually based kerning systems inherently better.

Algorithmic kerning systems can be remarkably effective, but ideally they should be used as a backup to traditional kerning-table systems. An improved system would frst look for specifc kerning data in the kerning table, and the information would be used if it were there. If the character pair in question weren’t listed in the font, the algorithm could have at it. Unfortunately, current systems offer an either-or proposition: You can have table-based kerning or algorithmic kerning, but the two will not work in concert. They will someday, but it hasn’t happened yet as this book goes to press.

Creating Custom Kerning Tables

If you’re unhappy with the quality of the kerning information in a font you use all the time, the frst thing you should do is try to find a replacement font. The first fonts created in the PostScript format, for example, were adapted directly from their dedicated-system counterparts, and these typically had very little kerning information built into them. It wasn’t unusual for early desktop fonts to list as few as 50 letter pairs in their kerning tables. It may be that a badly performing font is simply old and that a newer, better version of it exists.

An equivalent version of that font with better kerning information may also exist in the font library of another vendor. In either case, you will have to ask the foundry how many kerning pairs the font contains. There is no way to know this from the “outside” without a kern-table editor. As of this writing, the only page layout program that has one built in is QuarkXPress. The other option is to use a specialized font-editing program.

QuarkXPress allows you to edit the kerning tables for a font, altering existing values or adding new pairs and values of your own. The information is stored in a preferences fle that becomes part of the document, and it will be used every time that document is opened or printed. This procedure leaves your fonts in their original condition, which is a good idea.

In effect, using a font-editing program to alter kerning information creates a new font, or at least a unique version of an existing font. The problem with this is that the edited font is externally indistinguishable from the original. For copyright- and piracy-protection reasons, you can’t simply rename a font using a font-editing program. Creating a custom instance of a font, then, means having to keep strict control over it, because if it escapes into general circulation, it can cause untold composition problems, such as layout changes in documents based on the original version.

Kerning Numerals
The numerals in a typeface are generally lining figures with the same character width, and this works fine for nine of them. The 1, however, is clearly narrower than the others, with the result that it appears to set too loose. Only a handful of fonts offer kerning numerals (most of them in OpenType format); some contain an alternate 1 with a unique, narrower width. Old-style numerals in some fonts have unique character widths and may kern. When included in OpenType fonts as a layout feature, kerning numerals are referred to as proportional figures.

If your program allows you to create custom supplementary kerning tables for your fonts, you may want to create a set of kerning adjustments for the lining numeral 1 combined with the characters likely to set next to it, both before and after: the period (decimal point), hyphen, minus sign, en dash, comma, currency symbols, and the numerals 0 through 9.

If you choose to go this route, remember to turn off the automatic kerning for any numbers you set in tabular arrangements, because the kerned 1 will no longer align neatly in columns with the other numerals. It may be enough simply to kern the 1 manually in display situations only.

In the upper sample, the numeral 1 has been set with its native width, the same as that of all the other numerals. This makes it set rather loose. In the lower sample, hand-kerning has given the numbers in this headline a more natural and even spacing.

Using Tracking Controls

Tracking adjustments are most commonly applied to type as it grows in size. This is to counteract the impression that larger type has of setting looser than smaller type. Competent typesetting programs can adjust tracking automatically as point sizes change. Automatic tracking systems allow you to specify point-size thresholds at which tracking will automatically be tightened and by how much.

This tracking control panel can create a set of automatic tracking adjustments, which can be customized individually for every typeface you use. It offers five different tracks, from very loose to very tight, and each one has three point-size thresholds at which you can define specific tracking values. At sizes in between, the tracking value is interpolated, following the lines shown in the tracking editor illustrated here.

How much you should have the type’s tracking tighten automatically depends on the look you’re after. Some designers prefer very tight display type, while others prefer a looser setting. The illustration below shows a series of settings whose goal is to maintain the same sense of spacing from small to large.

Tightening the tracking as the point size grows can compensate for type’s tendency to look increasingly looser as its size increases. All these samples appear to have the same overall spacing, although in fact their tracking has been progressively tightened, with the largest sample being tightened by 50/1000 em.

Ideally, your program should allow you to set default tracking values as well as values for specifc typefaces (or it should set them for you). When you work with a specific typeface, you’ll learn how it needs to be tracked, if only because a particular client prefers a particular look. The default settings can be used as a starting point for tracking adjustments for other faces. Although the tracking demands of most individual faces don’t differ widely, they are different enough to demand your attention, as shown below.

There is no one tracking setting that works for all typefaces, as this illustration shows. The same setting that’s way too tight for Century Expanded (top) and only somewhat less so for Goudy Old Style (center) actually looks a little loose for Bodoni (bottom).

In addition to the automatic tracking values, programs typically give you the opportunity to add a supplementary tracking adjustment, so you don’t have to run off to the tracking editor every time the spacing of a passage of type has to be tweaked. In these supplementary controls, you typically specify the additional tracking adjustment in thousandths of an em, either as a negative value (to tighten the tracking) or a positive value (to loosen it).

Special Tracking Situations

One situation where you'll want to adjust tracking is when you are setting type on a patterned background, over screened line art or a photograph, or reversed out of a solid background. When type is set on a patterned background, it has to compete against the “noise” around it. Even type set on a solid, tinted background can have this problem, because when printed, the background will probably be screened—reduced into a fine array of dots—and the screening pattern will nibble at the edges of the characters. Loosening the tracking slightly opens up the spaces between the characters and makes them easier to read in these situations. Remember that type has been always designed as part of a black-on-white system.

In reversed type, loosening the tracking will also help with legibility in this unnatural white-on-black environment.

Low-resolution presentations also proft from looser tracking. The larger pixels in these displays make precise letter spacing impossible. As likely as not, two adjoining letters will kiss to create a single form, with pixels from one character leaning up against the pixels of its neighbors. Type set for use on websites, cd-rom manuals, e-books, and video presentations should be tracked looser than type destined for print.

For onscreen reading, loosening up the tracking improves readability. Even with the smoothed, anti-aliased type shown above, the characters in the normally tracked type at left often lean against each other. By print standards, the spacing of the right-hand sample is very loose, but onscreen it’s much more pleasant to read.

In general, adjusting the tracking as a copy-fitting technique is a bad idea. That is, avoid tightening the tracking to make too much type ft into not enough space. If you don’t have the ability to change the type specifcations (point size or leading, or both) or the page layout, get an editor involved. Too much copy is usually an editorial problem, not a design or typographic one.

Character Spacing and Script Faces
Script faces have very rigid spacing requirements, as almost all the characters are obliged to connect as they would in handwriting. You have to take care when adjusting the tracking of these faces to avoid disrupting these connections. Some script faces can tolerate some tightening of tracking, but few can have their tracking loosened much without their characters becoming detached from each other. Whenever possible, set script faces without any tracking adjustments at all.

In cases where letters set in a script face do not connect (such as between capitals and lowercase letters) you may have the opportunity to do some hand-kerning, but generally these faces are character-fitted very carefully and will need no such adjustments.

The top sample shows a script face with its normal, untracked spacing. Because of the way the characters overlap, this face can stand to have its tracking tightened somewhat (center), but if tracked too tight, the characters will overlap instead of merely meeting. The bottom sample shows that loosening the tracking too much can cause the connections between characters to become unstuck.

Text on Curved Baselines

In text set on curved baselines (usually called text on a path), characters with fat bottoms—especially wide characters—have a hard time. On a convex baseline (one that bulges upward) wide characters appear to teeter, while on a concave baseline (which sags downward) parts of those letters sink a little. Those letters are thus pulled out of vertical alignment with their neighbors. This happens to some extent on any curved baseline, but the effect becomes more exaggerated as curves become sharper.

Making the baseline visible shows how awkwardly the type sits on it. The I in the top sample looks too low because the neighboring characters have rocked up off the convex baseline. In the lower sample, the opposite is true, as the M and N now dip below the concave baseline, making the I appear to sit too high.

Narrower characters, then, fare better on curved baselines than wide ones—their narrow stance allows them to follow the contour better—so condensed faces typically work better in these settings. The sharper the curve, the more the width of the characters comes into play.

In addition, curved baselines change the spaces between characters from rectangles into wedges. This means that on a convex baseline the tops of the letters are spread apart, and on a concave baseline the tops of the letters are forced closer together. This is a kerning nightmare, because the familiar relationships among character shapes change. Some difficult pairs may be helped by having the baseline curved one way or the other, but others will become irremediably worse. The sharper the curve of the baseline, the worse the problem. Text set in lowercase fares better on curved baselines than all capitals because the relatively fewer ascenders reduce the gravity of the problems caused by those wedge-shaped letter spaces.

Nearly every character pair becomes a kerning problem when baselines bend. Comparing the type in the upper samples, you can see that what sets too loose on a convex baseline looks too tight on a concave one. In the lower samples, hand-kerning virtually every pair has resulted in relatively even spacing, although certain pairs, such as TV, remain intractable.

On convex baselines, some characters may have to be kerned close enough to touch each other along the baseline in order to bring some sense of logical spacing to their ascending parts. (This argues for using sans serif types in such situations, as you can space them tight without overlapping the serifs at their feet.) Conversely, type set on a concave baseline will have to be spaced farther apart to avoid collisions between ascending characters. In both cases, adjust the tracking first and tidy up with manual kerning (of which there will be a lot).

Text on convex baselines will usually be easier to read than that on concave baselines, but avoid setting type in a circular arrangement that puts some of the type upside down. The prescription for dealing with curved baselines, then, is threefold:

  • use gentle curves
  • choose condensed faces
  • hand-kern extensively

Don't miss the next Graphic Design article on Get the free newsletter in your mailbox each week. Click here to subscribe.

Excerpted with permission from The Complete Manual of Typography: A Guide to Setting Perfect Type, 2nd Edition (Adobe Press) by Jim Felici. Copyright © 2011.