Short answer
Give each category a hue that differs clearly from every other, keep lightness and chroma similar so no category looks more important than the rest, and stop adding colours at about six to eight. Beyond that, the smallest difference between any two colours shrinks until some pairs cannot be told apart, especially by readers with a colour-vision deficiency, and direct labels do more work than a longer palette.
Categories — countries, product lines, political parties — have no order, so their colours must not suggest one. ColorBrewer calls these qualitative schemes and describes them as relying mainly on differences in hue rather than lightness, precisely so the legend does not read as low-to-high. That gives a categorical palette two jobs that pull against each other: every colour must be different enough from every other to be matched to its legend entry at a glance, and none may be so much darker or more saturated that it looks like the answer. A palette of seven equally saturated mid-tones satisfies the second job; whether it satisfies the first depends on the smallest pairwise difference, not the average one.
Hue is a circle. Spread n colours evenly round it at one lightness and chroma and neighbouring colours sit 360/n degrees apart; at eight categories that is 45°, at twelve only 30°. The table below computes the smallest CIEDE2000 difference in such a set as n grows. For typical colour vision the worst pair falls steadily, from about 37 at three hues to about 11 at twelve. For simulated deuteranopia it is small at every count — even three evenly spaced hues include a near-identical pair — because a red–green deficiency folds the hue circle so that hues mirrored across the red–green axis land together, and at equal lightness nothing else separates them. The practical ceiling many chart libraries settle on — around eight to ten default series colours — is where the worst pair starts to become a guessing game even for typical vision. Past that, the better answer is fewer colours: group minor categories as 'other', facet into small multiples, or label lines directly.
Colourwise's palette checker treats ΔE00 below 10 as 'hard to tell apart at a glance for small chart marks'. That is an editorial threshold, much larger than the ΔE of 1–2 used for barely visible differences between large adjacent patches.
Holding lightness perfectly constant makes categories equal, but it also means the palette collapses in greyscale and relies entirely on hue, which is the dimension colour-vision deficiency removes. Designed categorical palettes therefore vary lightness deliberately. The Okabe–Ito set mixes a very light yellow, mid-tone orange and sky blue, and a dark blue, so pairs that a deuteranope confuses by hue are often still separated by lightness. The cost is that the yellow is almost invisible on white and the dark blue draws more attention than its neighbours; both need handling, usually by giving the pale colour an outline or keeping it for large areas rather than thin lines. The comparison table shows how this palette's worst pair under each simulated condition compares with seven evenly spaced rainbow hues.
A categorical colour should belong to the category for the life of a report. If 'France' is orange in one chart, it should be orange in the next, even if the second chart has only three series. Libraries that assign palette colours in data order break this silently: sort the data differently and every series changes colour. Fix the mapping explicitly — a dictionary from category to colour — and reserve particular colours for particular meanings across a publication. Where some categories matter more than others, it is better to use a categorical palette only for the few that matter and grey for the rest, which is covered in the emphasis topic, than to spend the most salient colours on unimportant series.
| Categories | Hue spacing | Worst-pair ΔE00, typical vision | Worst-pair ΔE00, deuteranopia (simulated) |
|---|---|---|---|
| 3 | 120.0° | 36.8 | 3.2 |
| 4 | 90.0° | 31.7 | 5.1 |
| 5 | 72.0° | 28.2 | 5.5 |
| 6 | 60.0° | 23.3 | 2.7 |
| 8 | 45.0° | 16.0 | 2.5 |
| 10 | 36.0° | 13.4 | 0.3 |
| 12 | 30.0° | 10.9 | 0.9 |
| Condition | Okabe–Ito (7 colours) | Evenly spaced rainbow (7 colours) |
|---|---|---|
| Typical colour vision | 21.7 (Orange / Yellow); 0 of 21 pairs under 10 | 6.3 (#33FF00 / #00FF66); 1 of 21 pairs under 10 |
| Protanopia (simulated) | 12.3 (Blue / Reddish purple); 0 of 21 pairs under 10 | 1.7 (#CCFF00 / #33FF00); 3 of 21 pairs under 10 |
| Deuteranopia (simulated) | 11.6 (Orange / Yellow); 0 of 21 pairs under 10 | 5.2 (#CCFF00 / #33FF00); 5 of 21 pairs under 10 |
| Tritanopia (simulated) | 7.8 (Orange / Reddish purple); 2 of 21 pairs under 10 | 1.4 (#33FF00 / #00FF66); 3 of 21 pairs under 10 |
| Greyscale print or display | 0.6 (Orange / Sky blue); 7 of 21 pairs under 10 | 0.0 (#33FF00 / #00FF66); 7 of 21 pairs under 10 |
Why: The charting library assigns palette colours by data order, and the data were sorted differently.
Fix: Map each category to a fixed colour explicitly and reuse that mapping everywhere.
Why: Twelve hues leave some neighbours too close for small marks, however well chosen.
Fix: Label lines at their ends, group minor series, or split into small multiples.
Why: Its colour is much darker or more saturated than its neighbours.
Fix: Even out lightness and chroma, or make that emphasis deliberate and explain it.
Each statement is labelled by kind — established fact, a standard’s requirement, observed market data, a convention, or Colourwise’s own interpretation or analysis — with the strength of the evidence behind it.
ConventionStrong evidence
ColorBrewer's qualitative schemes are intended for nominal or categorical data, do not imply magnitude differences between legend classes, and rely mainly on differences in hue.
Source: ColorBrewer 2.0 — colour advice for cartography; ColorBrewer.org: An Online Tool for Selecting Colour Schemes for Maps
ConventionModerate evidence
Okabe and Ito's colour-universal-design guidance recommends colours distinguishable across colour-vision types, combined with differences in shape, line type or position and with direct labels rather than colour-keyed legends.
Colourwise analysisModerate evidence
For hues evenly spaced round the OKLCH circle at fixed lightness and chroma, the smallest pairwise CIEDE2000 difference falls steadily as categories are added for typical vision, while under simulated deuteranopia even three hues include a nearly indistinguishable pair.
Based on: Computed by Colourwise with the Viénot–Brettel–Mollon dichromat simulation and CIEDE2000; see the hue-count table on this page.
Caveat: Simulation represents a full dichromat; most people with a colour-vision deficiency are anomalous trichromats and see larger differences than shown.
Source: Digital video colourmaps for checking the legibility of displays by dichromats; CIE (International Commission on Illumination) publications
Colourwise analysisModerate evidence
Seven evenly spaced rainbow hues keep far less separation between their closest pair than the Okabe–Ito palette under every simulated colour-vision deficiency.
Based on: Computed by Colourwise with the same analysis as the palette checker (Viénot–Brettel–Mollon simulation, CIEDE2000); see the comparison table.
Caveat: Both palettes have a weak pair somewhere; the Okabe–Ito set is not immune in greyscale, where orange and sky blue become almost identical.
Source: Digital video colourmaps for checking the legibility of displays by dichromats; Color Universal Design (CUD) — How to make figures and presentations that are friendly to colorblind people
Reviewed 29 September 2026. Colourwise summarises its sources in its own words and does not reproduce standards text or proprietary colour data. Spotted an error? Tell us.