Three color classes show only the roughest pattern; seven start revealing the finer structure

What it is

In a choropleth, the "grain" of the spatial pattern is set by the number of classes: a 3-class scheme compresses the map into three broad blocks and can only show the coarsest structure, a 7-class scheme reveals finer gradations and transitions, and an unclassed (continuous) scale preserves the most detail but makes it hard for readers to name a specific bin. The class count is a slider between pattern grain and readability: more classes means more detail but poorer discriminability between neighboring bins; fewer classes means a cleaner map but a compressed spatial structure.

Why it happens

The class count sits between two constraints on the visual channel. On the discrimination side, the number of reliably distinguishable color steps is limited: a sequential single-hue ramp is usually readable up to about 5-7 steps, beyond which adjacent bins become hard to tell apart on screen and readers start reading distinct bins as the same one. On the expression side, the class count determines pattern grain: few classes flatten internal differences (five distinct levels collapse into "medium"), while many classes expose local structure but amplify noise, because random fluctuation in small-sample regions is rendered as a "class difference". Combining both ends, the conventional guidance is 4-7 classes for a choropleth; use 3 only when the conclusion genuinely has three levels (pass / borderline / fail); above 8, either switch to a continuous scale or reconsider the chart type. The count also interacts with the number of regions: with a dozen provinces, many classes leave most bins empty, while across thousands of counties, few classes compress a large heterogeneous set into the same bin.

Where it stops holding

There is no single "correct" class count; it depends on the task. A lookup task ("which class is this county in?") favors fewer classes and a clean legend, while a pattern task ("where do high values cluster?") can benefit from more classes. The palette moves the ceiling: a single-hue lightness ramp becomes hard to read past 5 steps, a diverging palette can push discriminability to around 7, but neither is unbounded. Comparison constrains the choice too: two maps with different class counts cannot be compared color-to-color, so a map series must lock both the count and the breakpoints.

Applying it

• Default to 4-7 classes; adjust against the data histogram and task type, reserving 3 classes for conclusions that genuinely have three levels.
• Run a discriminability check after rendering: adjacent class swatches should be visually distinct; if they are not, reduce the count or change the palette.
• Lock the class count and breakpoints across a map series so colors stay comparable from map to map.
• Verify by having readers match a few regions back to the legend; the class count at which misreads climb is that map's discriminability ceiling—go below it.

Related

• Same group: U12.02.1 Large-area regions gain disproportionate visual weight in choropleth maps · U12.02.2 Choropleths should map rates rather than absolute counts · U12.02.3 Equal-interval, quantile, and natural-breaks classifications yield different map shapes · U12.02.5 Cartograms correct area bias at the cost of geographic recognizability
• Nearby: U1.09.1 Each visual channel supports only a limited number of reliably distinguishable levels · U12.02.3 Equal-interval, quantile, and natural-breaks classifications yield different map shapes
• Search terms: number of classes · color scale · choropleth design