Pick the visual that answers your question, and draw it so nobody misreads it.
The same figures can be read accurately or misread badly, depending on how they are drawn. This guide covers forty-six visual types, each with what it is for, when to use it and when to choose something else.
People judge position along a common scale most accurately, then length, then angle and area. That is why a bar chart usually beats a pie chart, and why area-based visuals need care.
Comparing categories, showing change over time, showing parts of a whole and showing a relationship are different jobs. Start with the question, then pick the visual.
Knowing when a chart stops working matters more than knowing what it is called, so each one names the case where you should choose something else.
Every visual links to its glossary term, where you will find the definition, why it matters and an explanation written for your role.
Sources: 1 Cleveland, W. S. and McGill, R. (1984), ‘Graphical Perception: Theory, Experimentation, and Application to the Development of Graphical Methods’, Journal of the American Statistical Association 79(387), 531–554, which ranks position, length, angle and area judgements by how accurately people make them. 2 Heer, J. and Bostock, M. (2010), ‘Crowdsourcing Graphical Perception’, ACM CHI, 203–212, which repeated those experiments at scale and found the same ordering. Recommendations here are ours, and the choices they support are judgement calls rather than rules.
The right visual depends on the question you are answering. Pick one to see the visuals that fit.
If you know your columns better than your intent, start here instead. Pick the shape that matches what you have, and the visuals that fit will show below.
Values for different categories as horizontal bars.
Values for categories as vertical columns.
Two or more series side by side for each category.
Values as a line ending in a dot.
Actual performance against a target and ranges.
Values as dots on a shared scale.
Exact values in rows and columns.
How a value changes across time.
A trend with the space below the line filled.
A tiny line chart without axes.
Change between two points in time.
Values that change at specific moments.
Totals broken into their parts.
Each category’s mix as percentages.
How a whole divides into slices.
A pie chart with space in the centre.
Parts of a whole as nested rectangles.
How increases and decreases build a total.
Values dropping through the stages of a process.
How often values fall into ranges.
The median, spread and outliers of data.
A smooth curve of where values cluster.
Each item plotted by two measures.
A scatter plot with a third measure as size.
Values as colour across a grid.
Columns and a line on shared or dual axes.
One key number with context.
Progress towards a target on a dial.
Values shaded across regions.
Flows between stages as bands of varying width.
Tasks shown as bars across a timeline.
Bars sorted largest to smallest with a running total line.
Scores across several measures shown as a shape on radiating axes.
A line with an average and upper and lower limits.
Box-plot detail with a smooth shape showing where values cluster.
Words sized by how often they appear.
A grid of small charts with the same scale, one per group.
A table grouped by rows and columns with subtotals.
A total broken down step by step into what drives it.
Bars extending either side of a centre line.
Stacked columns whose ribbons show how the ranking changes between periods.
A map with circles sized by value at each location.
A grid of squares where each square is a share of the whole.
Lines that track rank rather than value over time.
Rings that break a total down level by level, from the centre out.
Several key figures stacked in one block, each with its label.
Most charts that mislead are the right type, drawn carelessly. These nine rules cover where it usually goes wrong, with the same data drawn both ways.
A bar chart says "this one is twice as big" through its length. Cut the axis and a 4% difference looks like a doubling. Lines are different: they show change, so they may start near the data.
Alphabetical order scatters the story across the chart. Sorting by value puts the answer at the top. Keep the natural order only when one exists, such as months or survey scales.
Six colours make the reader work out what colour means before they can read anything. Use one colour, and a second only for the thing you want noticed.
A legend makes the eye travel back and forth matching colours to names. Put the name at the end of the line instead, and delete gridlines you do not need.
£1,284,309.47 is precision nobody acts on. Round to what changes the decision, keep the unit visible, and never mix per cent with percentage points.
The current week or month is always incomplete, so the last point falls off a cliff and people think something broke. Cut it, or mark it clearly as partial.
A gap in a line can mean no sales, or no data yet. Plotting blanks as zero invents a crash that never happened.
If the reader has to work out the point, half of them will work out a different point. One sentence in the corner settles it.
3D, shadows, gradients and heavy gridlines add ink without adding meaning, and 3D actively distorts the comparison.
Printable version: the cheat sheets download on the Templates page includes ten checks before you publish a dashboard.