Writing the paper

“The journal says my figures are too low-resolution.”

Formats, resolution, fonts, and colour-blind-safe colour for scientific figures.

The science is done. Then a submission system rejects your figures: too low-resolution, wrong format, fonts too small. Now you are exporting the same plot for the fifth time at midnight. The requirements read like a wall of numbers, but they come down to four decisions. This page settles each one, so your figures pass the check the first time.

By Heidi Andersén, MD, PhD · Updated 18 July 2026 · 8 min read

“I exported the figure at what looked fine on screen, and the system bounced it for being under 300 DPI. I have no idea what resolution my plot actually is.” A doctoral student, at the submission stage

The problemThe figure looks fine. The file does not.

A figure that looks crisp in your document can still fail a journal’s technical check, because your screen and the printed page are not measured the same way. On screen, a figure is a certain number of pixels. In print, it has to hold that detail inside a fixed physical width, one column or two, and that ratio, pixels over inches, is the resolution the submission system is testing. Paste a screen-sized plot into a two-column width and the resolution collapses.

The requirements are not arbitrary and they are not a test of taste. Every one of them, the format, the DPI, the font size, the colour, protects one thing: that a reader can see what you found, on paper and on a phone, whether or not they have typical colour vision. Read them that way and they stop being a wall of numbers. There are four.

Resolution is pixels over printed width. Fix the width, and the number is decided for you.

Common mistakesFour ways figures get bounced

Exporting a screenshot or a low-DPI PNG. A figure sized for a slide is far below print resolution. Blown up to a journal column, the lines go soft and the text furs at the edges.
Sending a bitmap where the journal wants vector. Line art (plots, diagrams, flowcharts) belongs in a vector format (EPS or PDF) that stays sharp at any size. Saving it as a flattened JPEG bakes in blur you cannot recover.
Type that survives on screen but not in print. Axis labels set to look right in your document are often 4-5 pt once the figure is scaled to column width: below the minimum, and unreadable.
Red-and-green as the only difference. Around one in twelve men cannot reliably tell your two conditions apart if red versus green is the only thing separating them. A rainbow (jet) colour scale has the same problem and adds false edges.

What they actually wantThe four decisions, in plain numbers

1 · Format. Match the format to the kind of image. Line art and plots, anything drawn from data, export as vector: EPS or PDF, which stay sharp at any size. Photographs, gels, micrographs, heatmaps export as a high-resolution bitmap: TIFF (preferred by most journals) or high-quality PNG. Some publishers accept only TIFF or EPS and will reject a JPEG outright, so check the guide for your target journal before you export. Whatever the format, embed the fonts.

2 · Resolution. Measured in DPI (dots per inch) at the final printed size, not at screen size. The widely used thresholds:

300 DPI: photographs and greyscale/colour images.
500-600 DPI: combination figures (a photo with lettering or line work over it).
1000-1200 DPI: pure line art saved as a bitmap. (Save it as vector instead and resolution stops being a worry.)

“At the final printed size” is the part that trips people. A single-column figure in many journals is about 89 mm (~3.5 in) wide and a double-column one about 183 mm (~7.2 in). Decide the width first, then export so the figure holds 300 DPI at that width: roughly 1050 px across for a single column, 2150 px for a double.

3 · Fonts. Use one clean sans-serif (Arial, Helvetica, or similar) and keep it consistent across every figure. After the figure is scaled to its column width, the smallest text should sit around 5-7 pt at the floor and no larger than the caption; roughly 8 pt is a safe working target for labels. Embed the fonts in the file so the journal’s system does not substitute them and shift your layout.

4 · Colour. Choose colours a colour-blind reader can still separate, and never let colour be the only signal. Avoid red-with-green pairings and rainbow/jet scales. Reach instead for a colour-blind-safe qualitative palette such as Okabe-Ito, or a perceptually uniform sequential map such as viridis for continuous data. Back colour up with a second cue (line style, shape, direct labels) and put the key inside the figure, not only in the caption.

A colour-blind-safe palette you can paste in (Okabe-Ito)

orange         #E69F00
sky blue       #56B4E9
bluish green   #009E73
yellow         #F0E442
blue           #0072B2
vermillion     #D55E00
reddish purple #CC79A7
black          #000000

Eight hues that stay distinguishable under the common forms of colour-vision deficiency. Use them in order, and pair each with a shape or line style so the figure still reads in greyscale.

Rather than paste the hexes, pull a tested palette from a package. In R, viridis gives perceptually uniform sequential maps, including cividis, which is tuned for colour-vision deficiency; khroma and ggokabeito provide the Okabe-Ito set above; and RColorBrewer flags which of its palettes are colour-blind-safe. In Python, viridis is matplotlib’s default.

If it fails in greyscale, colour was doing work it shouldn’t have to.

See itThe same plot, twice

Two groups over time. First in two of our brand violets; then recoloured for contrast with the Okabe-Ito palette and a second cue. Same data. Only the second reads for everyone, and it survives in greyscale.

Before: two close violets, one line style. Lovely on brand, but the series merge for a colour-blind reader and in greyscale.

After: Okabe-Ito blue and orange, one line solid and one dashed. Distinct by hue and by shape, so it holds even without colour.

ChecklistRun this before you upload

Six checks that catch most rejections

Format matches content: vector (EPS/PDF) for plots and line art, TIFF/high-quality PNG for photos and heatmaps, and the journal accepts it.
Resolution holds at the final width: 300 DPI photos, 500-600 combination, 1000+ line art if bitmapped.
Width set to the journal’s column (~89 mm single / ~183 mm double) before export.
Smallest text ~5-7 pt after scaling, one sans-serif, fonts embedded.
No red-green-only coding, no rainbow scale; colour-blind-safe palette, key inside the figure.
Print it in greyscale: if you can still read every series, colour is a bonus and not a crutch.