This entry is part of the Chart Index, the reference library for the Chart Design Field Guide.

Small multiples is not a chart type. It is a layout strategy: take any chart you would build for a single subject, and reproduce it as a grid of small panels, one per subject. Tufte called it the most powerful tool we have for visualising comparison. The principle is simple — same chart, same scale, different data — and the result is unmatched for exploring how a phenomenon varies across categories, regions, time periods, or experimental conditions.

The form is the antidote to the cluttered single-panel chart with twenty overlapping lines. Where a multi-line chart tangles, small multiples separate. Where a stacked bar with eighteen series stumbles, small multiples breathe. The cognitive trade is also clean: you are no longer reading one complex chart, but many simple ones, each easy to grasp.

What it is

Small multiples is a grid of charts, all of the same type, with the same axes and scale, each showing the same variables for a different subject. The grid is typically arranged by a categorical dimension (region, product, cohort). The reader's eye moves across the grid, picking up patterns by visual comparison: which panels are different from the others, in what way, by how much.

The hero chart above is twelve airports in a 4 × 3 grid, each panel the same line chart of monthly on-time departures for 2025. The y-axis is identical across panels — 60 to 95 per cent, ticked only on the first column — and the x-axis is shared, ticked only on the bottom row. Read across it and the shapes do the talking: the January dip that every airport shares, the deep July trough at Sydney and Melbourne that the smaller ports do not have, Hobart and Launceston sliding through the year while the Gold Coast climbs. An aggregated line chart of the same twelve series would have hidden all of it in a tangle. Randomise it and each panel jitters on its own, but the grid's scale holds, so every panel stays comparable with every other.

When to use it

Small multiples is the right choice when:

  • You have the same chart task for many subjects and want comparison across them.
  • The reader's question is "how does this pattern vary across categories?"
  • The single-panel alternative would be cluttered — too many overlapping lines, too many stacked bars.
  • The categorical dimension has a meaningful sort order — region, time, magnitude.
  • Each panel has enough room to be readable — typically at least 80 × 80 pixels per panel.
The hero's twelve airports on one panel
Twelve lines, twelve end labels, and a fifteen-point vertical range. The shared January dip is visible; nothing else is. Which airport had the bad July is a question this chart cannot answer without a hover.

When not to use it

  • Few subjects. Two or three panels are not small multiples; they are panels. Use a single chart with overlay or a slope chart.
  • Heterogeneous data per panel. If each subject has different variables or different scales, the form's comparison contract breaks. Use a dashboard of independent charts.
  • Very many subjects. 30+ panels become hard to scan. Filter or aggregate first.
  • When the inter-panel relationships matter. If you need to draw a line connecting subjects across panels (cross-panel comparisons), a single multi-line chart may be better.

Design principles

Use identical axes across panels

Every panel must share the same x-axis and y-axis ranges. The reading task is compare panels visually; that requires identical scales. The single most common mistake in small multiples is letting each panel auto-scale to its own data — this destroys comparability. The two figures below are the same six regions' monthly sign-ups; only the scale rule differs.

Independent scales: every panel fills its frame
Each panel is scaled to its own data, so all six lines look equally dramatic. Central, which grew by about twenty sign-ups, and Islands, which grew by about four, are drawn with the same sweep — and the tick labels on the left, which belong to the first panel in each row, now misdescribe every panel beside it. The eye is told these are comparable; they are not.
Shared scale: the same axis in every panel
The same six regions on one y-axis, 0 to 350. Now North's growth is visibly six times Central's, and Islands is a flat line at the bottom of its frame — which is the truth. Some panels will look empty on a shared scale; that emptiness is information.

Order panels meaningfully

Sort the grid by something that matters: total magnitude, geographic position, chronological order. Random or alphabetical order makes the chart a visual lottery; ordered, it becomes a story.

Nine stores in alphabetical order
Alphabetical is the order the spreadsheet came in. The three stores that are actually in trouble — falling all year — are scattered across the grid, and the reader has to find them one panel at a time.
The same nine stores, sorted by year-on-year change
Sorted from the biggest gain to the biggest loss, the grid reads top-left to bottom-right like a page. The three falling stores are now the bottom row, and the reader's eye lands on them without hunting.

Keep panels small but readable

The size sweet spot is 80–160 pixels per side. Smaller panels lose detail; larger panels start competing with each other for attention. Tufte's preference: as small as possible while remaining individually readable.

Use a single colour

Each panel uses the same colour — typically a single accent. Variation in colour across panels would imply a categorical encoding the data doesn't have. The point is the shape of each panel, not its colour.

One colour per panel
The same six regions with the palette cycled across panels. Nothing is encoded by the colours — the panel labels already say which region is which — but the eye reads six hues as six categories that want comparing by colour, and starts looking for a legend that does not exist.

Label each panel clearly

A short label per panel (region code, product name, cohort identifier) sits at the top of the panel. The label is the panel's identity; without it, the grid is anonymous.

Show only the axes that matter

Tufte's compact approach: only the left column carries y ticks and only the bottom row carries x ticks. The other panels inherit the scale; their axes are implicit in their position in the grid. This saves space and reduces non-data ink. Every grid on this page is drawn that way.

Annotate the panel that carries the story

If one panel has a notable feature (the only declining trend, the unexpected peak), call it out with a thin annotation or a coloured border. The grid provides context; the annotation provides narrative direction.

Anatomy

The composition of a small multiples grid
An anatomical guide

A small multiples grid's anatomy is a regular array of identical-shape panels, each carrying the same chart with different data, on the same scale, with minimal axis decoration. The form's reading is compare and contrast; the structure makes that comparison trivial.

  • Trellis / facet display — the statistical-software name for small multiples.
  • Single-panel multi-line chart — the cluttered alternative when the categories interrelate.
  • Horizon chart — small multiples compressed to strip height by folding the value range.
  • Stacked bar / area — for compositional comparison; less direct than per-panel comparison.
  • Heatmap — matrix-form alternative when colour-on-grid is preferred.
  • Slope chart panel — small multiples of slope charts; powerful for two-point cohort comparisons.

Reading list

  • Tufte, E. (1990). Envisioning Information. The foundational chapter on small multiples.
  • Becker, R., Cleveland, W. & Shyu, M. (1996). The Visual Design and Control of Trellis Display. The statistical formalisation.
  • Bertin, J. (1967). Semiology of Graphics. The conceptual roots of the small-multiple grid.
  • Wickham, H. (2009). ggplot2. The R implementation that popularised faceting in modern statistics.