How to Make Figures for Your Paper That Reviewers Actually Trust
Reviewers read your figures before they read your words — and a sloppy figure makes them distrust the whole paper. Here's how to build figures that carry the argument: one message per figure, axes that don't lie, color that survives grayscale and colorblindness, and the pre-submission pass that catches the mistakes that get papers desk-rejected.
1. Reviewers Read Your Figures First
Watch how an experienced reviewer actually reads a paper: title, abstract,
then straight to the figures. They form a hypothesis about your
contribution from the figures alone, and only then read the text to confirm
or refute it. By the time they reach your careful prose, they have already
decided whether they believe you. Your figures are not illustrations of the
argument — for the first pass, they are the argument.
2. One Figure, One Message
Before you make a figure, write the one sentence it is supposed to prove —
the takeaway a reader should be able to state after a five-second look. If
you can't write that sentence, the figure isn't ready; you have a data dump,
not a figure. A reader should never have to ask "what am I looking at?" or
"what am I supposed to notice?" The figure's design — what's big, what's
colored, what's centered — should answer both before they finish asking.
From data dump to argument
- Write the figure's one-sentence claim first; design backward from it
- If a panel doesn't support that claim, move it to the appendix
- The thing you want noticed should be the most visually prominent thing
- Six small panels of equal weight = no message; pick the hero panel
- The caption states the takeaway, not just 'Results for experiment 2'
3. The Craft That Keeps You Out of the Desk-Reject Pile
Most figure problems are not artistic — they are mechanical, and reviewers
notice every one. Text that's illegible because you exported at screen
resolution. Axis labels with no units. A legend covering the data. Fonts
that shrink to 5pt when the figure is placed in a two-column layout.
Default software styling that screams "I never touched the settings." None
of these require taste to fix; they require a checklist and ten extra minutes
per figure.
Mechanical fixes, ordered by how often reviewers flag them
- Label every axis with quantity AND units; no bare numbers
- Make text legible at final print size — ~8pt minimum in the printed column, not on your screen
- Export as vector (PDF/SVG) for plots; 300+ DPI for raster images
- Match fonts to the paper body; a Times paper with Arial figures looks bolted-on
- Move legends out of the data; or label series directly to skip the legend entirely
- Same color = same thing across every figure in the paper; keep a consistent palette
4. Color That Survives Grayscale and Colorblindness
Roughly 1 in 12 men has some form of color vision deficiency, and a
meaningful fraction of your readers will print your paper in black and
white or read it on a bad screen. A figure whose entire message lives in
the difference between red and green is invisible to all of them. Design so
that color is a bonus, not the only channel: distinguish your series by
shape, line style, and position too, so the figure still reads when color
is stripped away.
A palette that works for everyone
- Use a colorblind-safe palette (e.g. Okabe-Ito, or viridis for heatmaps)
- Never encode meaning in red-vs-green alone
- Add a second channel: solid/dashed lines, circle/triangle markers
- Print one copy in grayscale before submitting — can you still read it?
- Direct-label lines instead of relying on a color legend where you can
5. Honest Figures: Axes, Error Bars, and What n Was
The fastest way to lose a careful reader's trust is a figure that
technically isn't lying but is clearly arranged to mislead. A bar chart
with a y-axis starting at 85% turns a 2-point difference into a cliff. A
line plot with no error bars hides that your "improvement" is within noise.
A reviewer who catches one of these stops believing all your other
numbers, and you will spend the rebuttal digging out of a credibility hole
you dug yourself.
Honesty rules that protect your own credibility
- Start bar-chart y-axes at zero unless there's a stated, defensible reason
- Show variation: error bars, confidence bands, or individual points
- Say exactly what the error bars are — SD, SE, or 95% CI — in the caption
- Report n (seeds, runs, subjects) in or near every figure
- Don't cherry-pick the best run; show the distribution or the mean over seeds
6. The Pre-Submission Figure Pass
Do one dedicated pass over your figures the day before you submit, with the
manuscript compiled to its final two-column PDF — not the draft, not the
notebook. Read only the figures and captions, in order, as a skeptical
stranger would, and ask whether the story holds without the body text. This
is the single highest-return half hour in the whole submission, because it
catches the exact errors reviewers use to justify a reject.
The 10-minute pre-submission checklist
- Every figure referenced in the text, and every reference points to the right figure
- All text legible in the compiled PDF at 100% zoom — no 4pt tick labels
- Captions state findings and define every symbol, abbreviation, and error bar
- Consistent palette and notation across all figures
- Grayscale-readable and colorblind-safe
- Axes honest; n and variation shown
- Figures placed near their first mention, not all dumped at the end
PhD graduate who spent years tracking conference deadlines across computer science and engineering. Built ScholarDue after missing a submission window in the final year of candidacy and realizing no single tool tracked CFPs, extensions, and notification dates in one place.
Learn more→