Research Figure Design Principles That Make Your Data Impossible to Ignore

By SciScroll Team

Reviewers, editors, and readers decide within seconds whether a figure is worth their attention — and most figures fail that test not because the data is weak, but because the design buries it. Good research figure design is not decoration; it is the discipline of making your evidence legible at a glance to someone who will give it thirty seconds. This guide covers the principles that separate figures people skim past from figures people cite.

One Figure, One Message

Before you touch any plotting software, write the figure's message as a single sentence: "Knockout mice gained less weight than controls." If you cannot write that sentence, the figure is not ready. If you can, every design decision becomes a filter: does this element help a stranger understand that sentence faster? If not, cut it.

Two practical consequences:

Visual Hierarchy: Control What the Eye Hits First

The eye lands on the strongest contrast in the figure — the darkest color, the boldest line, the biggest deviation. Your job is to make sure that landing point is your actual finding, not an axis label or a decorative gridline.

Rank your elements deliberately:

  1. First: the data series that carries the message. Give it the most saturated color and the heaviest line weight.
  2. Second: direct labels on that series. A reader should grasp the finding without the legend.
  3. Third: context — baselines, controls, comparison series in muted grey or desaturated tones.
  4. Last (or never): gridlines, frames, tick clutter. Thin, light grey, or gone.

A quick self-test: blur your eyes (or shrink the figure to thumbnail size). If you can still tell what the figure is about, your hierarchy works. If everything dissolves into noise, your data is competing with your furniture.

Reduce Non-Data Ink

Every pixel that is not data costs attention. Chartjunk — heavy borders, shaded plot backgrounds, redundant legends, 3D effects, double-encoded decorations — forces readers to work for the payload.

A common before-and-after that fixes half of all manuscript figures: move the legend into the plot and drop the box border. The boxed legend outside the axes forces a ping-pong between key and data; direct labels next to each series eliminate the round trip. Then delete the outer frame (keep left and bottom axes), remove background fills, and thin gridlines to light grey.

Other quick wins:

For a systematic pass over these details before submission, work through a publication-ready figures checklist so nothing slips through.

Label Placement: Annotate, Don't Segregate

Labels exist to answer "what am I looking at?" at the point of contact. The further a label sits from its data, the harder the read.

Color: Semantics Across the Whole Manuscript

Readers experience the manuscript, not figures one at a time. If condition A is blue in Figure 1 and red in Figure 4, you are taxing every reader unnecessarily. Build a small palette — one color per condition — and use it everywhere: figures, schematics, even conference slides.

Three rules keep color honest:

If you are unsure which chart shape fits your data in the first place, the choice of chart type for research data interacts directly with color decisions — pie charts and stacked areas, for instance, are nearly impossible to label cleanly.

Design for Print, Not Just Your Monitor

Your figure looks great on a 27-inch display. It will be read at 8 cm wide, possibly in greyscale, by someone with low patience. Before submission:

A Concrete Before-and-After

Consider a typical "before": a five-series line plot in a full grey frame, boxed legend on the right, red and green series, gridlines at every tick, labels that shrink to 4 pt at column width.

The "after": frame reduced to left and bottom axes, gridlines deleted, colored labels at each line's endpoint, a blue/orange colorblind-safe pair, the key finding thickened to 2 pt while comparisons sit at 1 pt in grey, fonts at 7 pt final size. Same data — but the second version communicates its finding in three seconds instead of thirty.

Key takeaways

Frequently asked questions

What font size should research figures use?

Most journals expect figure text to be roughly 5–8 pt at final print size, but exact minimums vary by publisher, so check your target journal's author guidelines. The practical test: set the figure to its final column width on screen and confirm every label is readable without zooming.

Should figures be in color or black and white?

Design for both. Many readers print in greyscale, and some journals still charge for color in print. Use colorblind-safe palettes, distinguish series by shape or line style as well as hue, and convert your figures to greyscale before submission to confirm nothing disappears.

How many panels is too many for one figure?

If a figure needs more than about four panels, or if the panels do not support a single message, split it. A useful rule: each panel should earn its place by supporting the figure's one-sentence message. Supporting panels that only serve one section of the results often belong in supplementary material.

Do these principles apply to graphical abstracts too?

The underlying principles — one message, clear hierarchy, minimal ink — apply even more strongly, but graphical abstracts trade data density for a visual summary of the whole paper. If your journal requires one, our graphical abstract guide with template covers the differences in format and scope.

Want figures that look designed rather than defaulted? Browse the discover feed on SciScroll to see how other researchers present their data, then generate publication-ready scientific figures with AI assistance in the Studio — free to start, with paid plans if you need more (see pricing).