Pathform AI Browse packs

Research Figures

Research Figure Design Mistakes That Cost You Reader Attention

Figures are the part of a paper that readers see first and remember longest, which makes figure errors unusually expensive. A reader who cannot decode a figure does not simply skip it; they form a judgment about how carefully the rest of the work was done.

  • ClusterResearch Figures
  • Reading time5 min
  • FormatReview checklist
  • Sources5 listed

Guide clusterResearch Figures

Direct answer

The research figure mistakes that cost the most attention are unlabeled axes, type that fails at print size, color that disappears in grayscale, captions that only name the figure, and elements that imply relationships the data do not support. Each has a concrete fix and most can be corrected in one editing pass.

01Errors in labeling and units

Unlabeled axes are the most common figure fault in submitted manuscripts and the easiest to fix. Every axis needs a label with its unit, and every quantity needs to be unambiguous: is the reported value a mean, a median or a fitted coefficient, and over what population. Where error bars appear, the caption must say what they represent, because conventions differ across fields and a reader cannot infer the choice.

Related errors include inconsistent significant figures, units mixed within one panel, and categorical labels that shift between the figure and the table. These are small individually, but each one costs the reader a moment of doubt, and accumulated doubt is what makes a paper tiring to read.

  1. Label every axis with the quantity and its unit.

  2. State in the caption what error bars, shading or bands represent.

  3. Keep decimal precision consistent and justified by the measurement's resolution.

  4. Check that category labels match the tables and the text exactly.

02Errors in legibility

Figures are usually built at a comfortable size and then reduced by a journal's layout, sometimes to half or a third of the original width. Text that was legible in the drafting tool becomes unreadable at final size. The reliable practice is to set type at what will be its final size after reduction, then design around it rather than after it.

Line weights fail in the same way. Hairlines that render crisply on a high-resolution screen disappear in print or on lower-resolution displays. Symbols need to remain distinguishable when small, which usually means larger markers and stronger contrast than feels necessary during drafting.

ElementTypical failurePractical target
Axis labelsToo small after journal reductionSet at final printed size, not drafting size
Line weightsHairlines vanish in printUse visible weights and test in grayscale
MarkersSymbols become indistinguishable when smallIncrease size and separate shapes clearly
Panel lettersInconsistent size and placementUse one size and one corner position throughout

03Errors in color and accessibility

Color is the most common carrier of meaning in research figures and the least reliable. Roughly one in twelve men has some form of color vision deficiency, and many readers print in grayscale. A figure that distinguishes two conditions only by red and green becomes unreadable for those readers, which is an accessibility failure and, for many journals, a compliance issue.

Use position, shape, pattern or direct labeling alongside color, and check every figure in grayscale before submission. Sequential and diverging color scales should be used according to the data: sequential scales for ordered magnitude, diverging only where there is a meaningful midpoint. Rainbow scales distort perceived differences and should be avoided for continuous data.

If the figure only works in color, it does not work.

04Errors in captions

Captions are frequently written last and treated as labels. A good caption does three things: it states what the figure shows, it defines any symbols or abbreviations used, and it points to what the reader should notice. That third element is what distinguishes a caption that helps from one that merely identifies.

Captions should also be self-contained, because readers often encounter figures out of context, in a search result, in a slide or in a shared image. If the caption cannot be understood without the surrounding text, it will be misunderstood by everyone who sees the figure separately, which is a larger group than most authors expect.

  • State what the figure shows in the first sentence.
  • Define every symbol, abbreviation and error bar type.
  • Point out the comparison or pattern the reader should notice.
  • Keep the caption free of interpretation that belongs in the results text.

05Errors that cross into misrepresentation

Some figure choices do not merely obscure data, they distort it. Truncated axes exaggerate differences and should be avoided for bar charts, where the zero baseline is expected. Dual axes can create apparent correlations that reflect chosen scaling rather than data. Overlapping or three-dimensional elements can hide the very comparisons the figure is meant to support.

Where a chart form requires a non-zero baseline, such as a line chart of a change over time, say so explicitly in the caption and choose the range for a stated reason. Transparency about scaling choices protects both the reader and the author, because a reviewer who suspects an axis was chosen to flatter the result will read the entire analysis more skeptically.

Worked example

Worked example: repairing a figure at revision

A paper reported that a new catalyst increased reaction yield relative to a control across five temperature conditions. The submitted figure used a bar chart with a vertical axis starting at seventy percent, red and green bars to distinguish the two conditions, no error bars, and a caption that read simply, yield by temperature.

The reviewer's objection had two parts. The truncated axis visually exaggerated a difference that the text described as modest, and the color scheme was the only way to distinguish conditions, which made the figure unreadable in grayscale and for readers with color vision deficiency.

The revision addressed both. The axis was reset to start at zero, with the range justified by the fact that a yield of zero is physically meaningful. Conditions were distinguished by fill pattern and direct labeling of the two series at the right-hand end of the chart, with color used only as a redundant cue. Error bars were added from the replicate measurements, and the caption now stated that the bars represented the standard deviation across three independent runs.

The revised caption also pointed to the comparison rather than merely naming the figure: the gap between conditions narrowed as temperature increased, which was the paper's actual claim. The numerical difference the authors wanted to emphasize was conveyed in the text, where it could be stated with its uncertainty rather than implied by axis scaling.

Common mistakes

Unlabeled axes or undefined error barsLabel every axis with quantity and unit, and state in the caption what error bars represent.
Type and lines sized for the drafting viewSet type and line weights at the final printed size, then check in grayscale.
Color as the only distinguishing cueAdd pattern, shape or direct labeling so the figure works in grayscale and for color-blind readers.
Truncated axes that exaggerate a differenceStart bar charts at zero, and where a non-zero baseline is necessary, justify the range in the caption.
  • Research Architecture Pack
  • Research Figure Pack
  • Technical Roadmap Pack
  • How to Design the Central Diagram for a Research Poster
  • Research Figure Types: Choosing the Right One for Your Message
  • Methodology Diagram Mistakes That Reviewers Notice First

Frequently asked questions

Should error bars show standard deviation or standard error?

Either can be appropriate, but the choice must be stated in the caption, and consistency matters more than the convention. Many journals now prefer confidence intervals, which convey uncertainty on the scale of the estimate. Whatever you choose, do not leave it to the reader to guess.

How do I make a figure accessible without making it ugly?

Use direct labels instead of legends, choose colorblind-safe palettes, add pattern or shape as a secondary cue, and keep contrast high. These choices usually improve clarity for all readers, not only those with color vision deficiency, so accessibility and quality rarely conflict.

Can an AI tool check my figure for these problems?

It can review a description of your figure and flag missing axis labels, undefined error bars or color-only encoding, which is a useful pre-submission check. It cannot see whether the data support the visual comparison, so the substantive checks remain yours.

Do I need to redraw all figures to a single style?

Consistency across a paper helps readers transfer conventions from one figure to the next, particularly for colors that encode the same category. Perfect uniformity is not required, but a reader should never have to relearn what a color means.

Sources and further reading

  1. Author services - submission and figure guidanceWiley
  2. JATS - Journal Article Tag Suite, the standard article structureNational Center for Biotechnology Information, U.S. National Library of Medicine
  3. ColorBrewer - colour schemes tested for print and screen legibilityCynthia Brewer, Pennsylvania State University
  4. WCAG 2.1 at a glance - contrast and non-text accessibility requirementsWorld Wide Web Consortium (W3C)
  5. Information for authors - artwork and submissionElsevier

These are the organisations whose published guidance the recommendations on this page follow. Listing them does not imply that they endorse Pathform AI, and this page is not affiliated with them.

Put this guide into practice

Get the editable layouts this guide describes, ready to retype and submit.

Browse template packs