How to make a publication-ready figure

Before opening any plotting software, write one sentence that says what the figure should show: "Treatment B lowered blood pressure more than A at every time point" or "Yield rises with temperature up to 60 °C, then falls". That sentence decides the chart type, what goes on each axis, and what to emphasise. A figure that tries to show everything usually shows nothing clearly.

Show the data, not just the summary#

Bar charts of means with error bars (sometimes called dynamite plots) hide the sample size, the distribution and any outliers. Very different datasets can produce the same bars. For small to moderate samples, plot the individual observations, with the mean or median and an interval on top. Several journals now encourage this.

Design the figure at the size it will be printed. Journals usually specify a single-column width (often around 85–90 mm) and a double-column width (often around 170–190 mm), plus a maximum height. Check yours.

• Set the figure size in your software to the final width before adding text, rather than shrinking a large figure later. Shrinking makes text and lines too small.
• Use one sans-serif font throughout, at sizes that remain legible when printed. Many journals state a minimum font size; follow it.
• Keep line weights consistent and thick enough to survive reduction.
• Remove what does not carry information: heavy gridlines, background shading, borders, 3D effects and redundant legends.
• Label panels consistently (A, B, C) in the position and style your journal asks for.

• Every axis gets a label with units in a consistent format, for example "Temperature (°C)" or "Reaction time (ms)".
• Start bar chart axes at zero, because bar length encodes the value. For dot plots and line plots a non-zero baseline is acceptable when it is clearly marked.
• Use the same axis range for panels that readers will compare.
• Label lines and groups directly where you can, instead of making the reader match colours to a legend.
• Say exactly what error bars show (standard deviation, standard error or 95% confidence interval) and give n for each group, in the caption. They mean very different things.

Colour vision deficiency is common, particularly among men, and many readers will print your paper in greyscale. So:

• Do not rely on red versus green alone to distinguish groups.
• Use a palette designed for colour-blind readers, such as the Okabe–Ito palette for categories, or viridis-family colour maps for continuous values.
• Avoid rainbow colour maps (such as "jet") for continuous data: they create apparent boundaries that are not in the data and do not convert well to greyscale.
• Back colour up with a second cue (shape, line style, direct labels), and check the figure in greyscale or with a colour-blindness simulator.
• Use colour to highlight the comparison that matters and grey for context.

Resolution only means something at a given physical size. A 3000-pixel-wide image is 300 dpi at about 254 mm wide, but only about 150 dpi if printed at twice that width. Export at the final size.

A reader should understand the figure from the caption without searching the text. A good caption has:

1. a title sentence stating what the figure shows (some styles ask for the main finding here),
2. what each panel, symbol, colour and line represents,
3. what the error bars or bands show, and the sample size,
4. any statistical test whose result is marked on the figure, and what the markers mean,
5. abbreviations spelled out, and the source if the figure is adapted from someone else's work (with permission where needed).

Caption and numbering conventions differ between styles. APA 7, for example, puts the figure number and title above the figure. See the APA 7 and IEEE pages, or your journal's guide.

• Generate figures with a script or saved workflow, not by hand, so you can regenerate them when the data change.
• Keep the raw data and unprocessed images. Journals may ask for them.
• For images, apply adjustments such as brightness and contrast to the whole image, never to selected regions, and describe them in the Methods. Do not splice lanes or panels together without marking the join.
• Write alt text that describes what the figure shows for readers using screen readers, if your venue supports it.

Before submitting, run through your target journal's figure checklist, and for a thesis, the thesis formatting checklist. If the figure reports a two-group comparison, our guide on choosing a statistical test covers what the caption should report. You can see figures in context on the lab reports use case.

Print the figure at its final size, or view it at 100% zoom at that size, and check:

• Can you read every label without zooming?
• Does it make the point in your one-sentence summary within a few seconds?
• Are units, n and the meaning of error bars stated?
• Do colours still work in greyscale, and is nothing distinguished by red versus green alone?
• Are fonts, line weights and panel labels consistent with your other figures?
• Does the file format and resolution meet the journal's or institution's requirements?
• Is the figure numbered, cited in the text, and in the order it is first mentioned?

If any answer is no, fix it now. Reviewers and examiners notice figures first, and a clear figure makes the rest of the paper easier to trust.