The three types of data visualization color scales
Every chart color decision belongs to one of three scale types. Sequential scales encode ordered, one-directional data — temperature, revenue, density. They work best with a single hue progressing from light (low values) to dark (high values). Diverging scales encode data with a meaningful midpoint — profit/loss, above/below average, agree/disagree. They use two contrasting hues that meet at a neutral center, allowing the viewer to read both direction and magnitude from color alone. Categorical scales encode unordered group membership — product lines, regions, demographic segments. They use maximally distinct hues at equal perceptual weight, so no category appears more important than others simply because of its color.
Why rainbow color scales are a mistake
Rainbow (ROYGBIV) scales appear to encode a range, but they fail in four ways. First, perceived brightness is uneven: yellow is far lighter than blue, creating false visual emphasis in the middle of the range. Second, there is no logical ordering — a viewer cannot reliably determine whether violet is above or below red without a legend. Third, the hue steps are perceptually unequal: the gap between red and orange appears larger than the gap between green and teal even if the data gap is identical. Fourth, the red-to-green transition is the most common color-blind confusion pair, making the scale unreadable for the 8% of men with deuteranopia. Use perceptually uniform alternatives: viridis (blue-green-yellow), cividis (blue-yellow, fully color-blind safe), or plasma (blue-pink-yellow).
Building accessible categorical palettes
A categorical palette for charts should meet three criteria. (1) Sufficient hue separation: each category hue should be at least 30° apart on the color wheel. A practical starting palette: blue (220°), orange (25°), green (140°), red (0°), purple (280°), teal (175°). (2) Equal perceptual weight: saturation and lightness should be roughly equal across all hues so no category visually dominates. (3) Color-blind legibility: test every palette with a deuteranopia simulator. The blue-orange combination is the most reliably accessible pair, as it is distinguishable for all common types of color vision deficiency.
Color and chart type: matching the scale to the visualization
Different chart types have different color requirements. In line charts, each line needs a distinct categorical hue — limit to 5-6 lines before switching to direct data labels that remove the need for color-based differentiation. In bar charts, bars within a single category should all share one hue; adding color variation implies categorical differences that may not exist. In scatter plots, color encodes a third variable — apply a sequential scale for quantitative third variables, a categorical scale for nominal ones. In heatmaps, sequential scales (light to dark) are almost always correct; diverging scales are only appropriate when the data has a genuine neutral midpoint with meaningful departures in both directions.
Never use color as the sole differentiator
The most important rule in accessible data visualization is that color should never be the only way to distinguish between data categories or values. Every color-encoded distinction should have a secondary encoding: position, shape, pattern, direct label, or texture. This is not just a color-blindness consideration — it benefits all users in high-glare environments, on printed materials, in black-and-white photocopies, and in screenshots shared in document form. The principle is: design with color, verify without it.