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

The pie chart is the form everyone has used and almost no expert recommends. Tufte called it a chart that should never be used. Few softens this to almost never. The objection is perceptual: the eye is poor at comparing angles and areas, and the pie chart asks it to do both. For nearly every comparison task, a bar chart is more accurate, more compact, and easier to read.

Yet pie charts persist — because audiences understand them, because they communicate the part-of-a-whole concept instantly, and because, used within their narrow band of competence, they are not always wrong. The donut is the same chart with the centre removed: a small concession to data-ink that does nothing for the underlying perceptual problem.

What it is

A pie chart maps a categorical variable to slice colour and a quantitative variable to angle. The full circle represents the whole; each slice is a fraction of the whole proportional to its value. The donut variant removes a central disc, leaving an annulus — visually a ring rather than a disc.

The hero chart above is the form within its comfort zone: five sources of electricity, one of which is nearly half the whole. The reader sees at once that coal still dominates, that the three renewable sources together are close to matching it, and that gas and hydro are minor. Those are part-to-whole statements, and the pie makes them without effort. Try ranking solar against wind by angle alone and the trouble starts: the labels do the work the wedges cannot. Randomise the values and watch how hard it is to say which of the smaller slices grew.

When to use it

Pie charts are defensible when:

  • You have two to five categories that sum to a meaningful whole.
  • There is one dominant slice whose dominance is the entire point.
  • The audience expects the form and would find a bar chart unfamiliar (board packs, executive summaries, public communication).
  • Precise comparison between slices is not required — only the part-to-whole relationship.

That list is short, and most analytical work falls outside it. When in doubt, use a bar chart.

When not to use it

  • More than five slices. Six or more slices fragment into a wheel of similar-sized wedges that cannot be ranked by eye. Switch to a horizontal bar chart, sorted.
  • Slices of similar size. Three slices of 32%, 34%, and 34% are indistinguishable as wedges. A bar chart makes the difference visible.
  • Comparing two pies. Did the mix shift between Q1 and Q2? requires comparing wedge by wedge across two charts, which the eye cannot do. Use a stacked bar or slope chart.
  • Negative values, zero values, or anything that does not sum to 100%. The form depends on parts summing to a whole. Break that and the chart becomes meaningless.
  • Tiny slices that matter. A 2% slice may be the most important value in the dataset and is unreadable in a pie. Use a bar chart and let the eye see it.

Design principles

Limit to five slices

Three to five slices is the comfort zone. If you have more categories, rank by size and consolidate the long tail into Other. This is the most important rule and the one most often broken. The two donuts below hold the same sessions; only the number of slices differs.

Twelve slices: a wheel that cannot be ranked
Sessions by traffic source, every source its own slice. Once the eye leaves the largest two wedges it has no way of ordering the rest, and half the labels are fighting for the same patch of margin.
The same sessions as five slices
The four largest sources kept, the eight smallest folded into Other, and the slices drawn largest-first from 12 o'clock. The chart now answers the only question a pie can answer: roughly how much of the whole does each big source hold?

Sort by size, starting at 12 o'clock

The eye expects to start at the top and move clockwise. Start the largest slice at 12 o'clock, then descend by size around the circle. Leave the Other slice last, as in the figure above. This convention reduces the work of reading and ranking.

Prefer position when the values are close

Angle comparison fails first when slices are nearly equal. The pie below has three slices within two points of each other; the bar chart underneath holds exactly the same numbers.

Three near-equal slices: which is largest?
Share of support tickets by channel. The wedges are 35, 33 and 32 percent, and without the printed values nobody could put them in order. Randomise the values and notice that the picture barely seems to change even when the ranking does.
The same three values as bars
Position along a common scale. The two-point gap between chat and email is now a visible difference in length, and the order reads top to bottom without a single value being consulted.

Never ask the reader to compare two pies

A mix that changes over time is the commonest reason people draw pies in pairs, and it is the comparison the form handles worst: every wedge has moved, so nothing lines up. Put the shares on a slope chart instead and the change is the picture.

Channel mix, Q1 to Q2, as a slope chart
Each line is one traffic source's share of sessions. Two pies would have asked the eye to match wedges across charts; here paid search's fall and email's rise are read from the slopes alone, and the crossing tells you email overtook paid search.

Direct-label, with values

Place each label and percentage at or near its slice, not in a separate legend. The reader should never have to match colours between a key and a wedge. If the slices are too small to label inside, lead lines to outside labels — but legends are a last resort. Every pie in this entry prints the value inside any slice wide enough to hold it and hangs the category name on a short leader outside.

Quiet the palette

Pie charts magnify colour. Saturated rainbow palettes turn the form into decoration. Use desaturated, related hues from the categorical palette; if there is a hero slice, give it the accent colour and let the others recede. The five-slice donut above gives Other the neutral grey precisely so it reads as a remainder rather than a category.

Donuts add little

Removing the central disc reduces non-data ink by a small amount and provides space for a centred total label. That is its only real benefit. It does not improve angle comparison; if anything, the arc length further removes the eye from the central pivot that helps anchor angle estimates.

Avoid 3D pies

3D pies distort the apparent size of every slice — front slices read larger than rear slices of identical value. They are a perfect example of chartjunk: ornament that actively misleads. Refuse to ship them.

Show the total

Whether pie or donut, the whole deserves a clean label. Total revenue · $186m FY25 — placed centrally for a donut, or as a subtitle for a pie. Without the total, the percentages float without anchor.

Anatomy

The composition of a donut chart
An anatomical guide

The structural elements of a pie chart are minimal: slices, labels, a total. Everything else — borders, drop shadows, exploded slices, 3D perspectives — is chartjunk in Tufte's strict sense.

  • Bar chart — the perceptually superior alternative for nearly every comparison the pie attempts.
  • Stacked bar — for part-to-whole when you want to compare two or more compositions side by side.
  • Slope chart — for a mix that changes between two periods; the honest replacement for a pair of pies.
  • Waffle chart — a 10×10 grid where each cell is 1%. Reads as part-to-whole without the angle problem.
  • Treemap — for part-to-whole when you have many categories and hierarchy matters. Covered in a separate entry.
  • Donut — the pie's twin, with a centre cutout for a total label.

Reading list

  • Tufte, E. (2001). The Visual Display of Quantitative Information. The original objection to the pie chart.
  • Few, S. (2007). Save the Pies for Dessert. The definitive critique with a narrow defence.
  • Cleveland, W. & McGill, R. (1984). Graphical Perception. The perceptual hierarchy that places angle at the bottom.
  • Spence, I. (2005). No Humble Pie. The historical and cognitive case for the form.