Designing Scientific Figures That Make the Result Clear

A scientific figure is part of the evidence, not decoration added after the analysis. Its design determines which patterns readers can inspect, which comparisons appear important and whether uncertainty remains visible.

A visually polished figure can still be scientifically weak. Data may be compressed into summaries that conceal their distribution, colour may create differences that are not present in the measurements, or multiple panels may be arranged without a clear analytical sequence. Figure design should therefore begin with the scientific question rather than the plotting software.

Define the task of each figure

Before selecting a chart type, write one sentence describing what the figure must allow the reader to evaluate. For example:

These tasks require different visual forms. A chart chosen because it is familiar or aesthetically attractive may not expose the comparison that matters.

Each main figure should also have a defined role in the manuscript. If its contribution cannot be stated separately from another figure, the two may be redundant or may need to be combined.

Choose the representation from the data

The structure of the dataset should determine the graphic. Continuous measurements, counts, proportions, time series, paired observations and spatial data do not carry the same information.

1. For distributions, show the observations. Dot plots, violin plots, box plots or combinations of these reveal dispersion and possible outliers more effectively than bars containing only a mean and error bar.
2. For relationships, preserve both variables. Scatter plots allow readers to inspect form, variation, clustering and observations that strongly influence a fitted relationship.
3. For repeated or paired measurements, retain the pairing. Connecting observations from the same experimental unit can expose changes that are hidden when groups are summarized independently.
4. For time series, show temporal resolution honestly. Lines imply continuity and order. They should not be used to connect unrelated categories.
5. For images, provide quantitative context. Representative micrographs should be accompanied by scale information, clearly defined processing and appropriate quantification.

Summary statistics remain useful, but they should not replace the empirical structure of a small or heterogeneous dataset.

Keep the scale honest

Axis choices affect the apparent magnitude of an effect. A restricted range may be appropriate when small changes are scientifically meaningful, but the restriction must be visible and justified. Truncated axes are particularly problematic in bar charts, where length is interpreted relative to a baseline.

Logarithmic transformations should be indicated clearly and used because they suit the data or analysis, not because they produce a more favourable visual separation. Comparisons across panels are easier when related plots use the same range, units and aspect ratio.

Check that:

• units appear on every quantitative axis;
• tick spacing is regular and interpretable;
• related panels use consistent scales where possible;
• transformations are identified in the label or legend; and
• reference lines and thresholds have a defined scientific meaning.

Show variation and uncertainty

A central estimate without information about variability gives an incomplete account of the result. Readers should be able to determine whether error bars show a standard deviation, standard error, confidence interval or another quantity.

Sample size should be stated, and the experimental unit must be clear. Multiple measurements from the same sample are not equivalent to independent biological or observational replicates. A figure can appear densely populated while resting on few independent units.

When model estimates are shown, display uncertainty around the estimate where appropriate. For fitted relationships, this may take the form of confidence bands. For effect estimates, intervals are often more informative than significance symbols alone.

Use colour to encode information

Colour should have a defined function. It may identify experimental groups, represent an ordered quantity or direct attention to a specific result. Additional colours that do not encode information increase visual complexity without improving interpretation.

Select palettes according to the structure of the data:

• Categorical palettes distinguish groups without implying an order.
• Sequential palettes represent values progressing from low to high.
• Diverging palettes show movement in two directions around a meaningful reference point.

Avoid rainbow scales and red–green contrasts. Uneven changes in brightness can create artificial boundaries or conceal genuine variation, and some combinations are inaccessible to readers with colour-vision deficiencies.

Colour should not be the only means of distinction. Symbols, line styles, direct labels and panel separation can preserve meaning when figures are printed in greyscale or viewed under imperfect conditions.

Remove competition between the data and the design

Heavy borders, background shading, unnecessary grid lines, three-dimensional effects and decorative symbols compete with the observations. Most scientific figures benefit from fewer visual elements.

Reduction should not remove necessary context. Scale bars, control groups, sample identifiers, uncertainty and analytical thresholds are not clutter when they are required to interpret the evidence.

Use a consistent visual system across the manuscript:

• assign the same colour to the same group in every figure;
• use consistent fonts and label sizes;
• place panel letters in the same position;
• use the same terminology and abbreviations; and
• align panels to a common grid.

Build multi-panel figures as an argument

Panel order should follow the reasoning of the experiment. A reader will usually move from the experimental premise to the principal observation, supporting analysis and, finally, a mechanistic or integrative result.

Do not assume that alphabetical ordering creates a logical sequence. Ask whether panel B answers a question raised by panel A and whether panel C is needed to interpret the comparison that precedes it.

Panels should be large enough to read at publication size. Combining many plots into one figure may save nominal space while making every component difficult to evaluate. Move secondary analyses to supplementary material only when they are not necessary to establish the main claim.

Write legends that define the evidence

A figure legend should allow the graphic to be understood without searching through several sections of the manuscript. It need not repeat the results, but it should define what was measured and how the visual elements should be read.

A complete legend generally identifies:

• the experiment or analysis represented;
• the biological system, population or dataset;
• the meaning of symbols, colours, lines and shaded regions;
• the number and nature of independent observations;
• the summary statistics and error measures;
• the statistical test and any correction for multiple comparisons; and
• relevant scale bars, thresholds or image-processing steps.

Avoid vague statements such as “data are shown as mean ± error.” Name the error measure and state what the sample size represents.

Preserve image integrity

Adjustments to scientific images must not alter the interpretation of the data. Brightness or contrast changes should normally be applied to the entire image and should not remove, add or selectively enhance features.

Cropping may be necessary, but it should not exclude relevant context or create a misleading comparison. Images shown side by side should be processed consistently unless differences in processing are explicitly explained.

Retain the original files and a record of processing steps. For composite images, boundaries between samples, fields or experimental conditions should be visible.

Test the final figure under realistic conditions

A figure that works on a large monitor may fail when reduced to journal-column width. Export the complete figure at its intended size and inspect it without zooming.

Then test it again:

• in greyscale;
• with a colour-vision-deficiency simulator;
• on a standard laptop display;
• as a printed page, if print use is likely; and
• without the caption, to identify missing labels or context.

Ask a colleague who was not involved in the analysis to describe the principal comparison. If their interpretation differs from the intended one, the figure needs revision.

Final figure check

• Every figure has one defined scientific task.
• The chart type matches the structure of the data.
• Individual observations are shown where they aid interpretation.
• Axes, units, transformations and reference points are explicit.
• Sample sizes and experimental units are defined.
• Variation and uncertainty are represented appropriately.
• Colour encodes information and remains accessible.
• Related groups retain consistent visual identities.
• Multi-panel figures follow the scientific argument.
• Legends define symbols, statistics and experimental context.
• Image processing has not altered the scientific interpretation.
• All elements remain readable at the intended publication size.