Here is something easy to underestimate: a reviewer's first impression of your paper is visual. Before anyone reads your methods, they scan your figures and tables. If those are cluttered, blurry, or duplicate each other, the paper feels weaker before a single argument has been judged. Getting your display items right is some of the highest-return work you can do on a manuscript.
Figure or table? Choose deliberately
The two are not interchangeable. A figure is for a pattern — a trend over time, a difference between groups, a relationship between variables that the eye can grasp instantly. A table is for precise values a reader may want to look up or compare exactly. The cardinal sin is presenting the same numbers as both a figure and a table; pick the one that serves the reader and delete the other.
Resolution and file formats
Low-resolution images hold up production and cannot be fixed by the journal. Prepare each figure at the size it will be printed, and at these minimums:
| Figure type | Minimum resolution | Preferred format |
|---|---|---|
| Photographs / micrographs | 300 dpi | TIFF or high-quality JPEG |
| Graphs, charts, line art | 600 dpi | Vector (EPS/PDF/SVG) or high-res PNG |
| Fine line drawings | 1200 dpi | Vector |
Wherever possible, submit graphs as vector files. Vectors stay razor-sharp at any size, so a chart exported as EPS or PDF will never pixelate the way a screenshot does. Enlarging a low-resolution image never adds detail — it only makes the blur bigger.
Treat the values above as a working rule of thumb, not a universal standard. The target journal's author guidelines set the real requirement, so check them before you export and follow them where they differ.
Designing a figure people can read
- One message per figure. If a figure is trying to say three things, it will say none of them clearly. Split it into panels (a, b, c) with a shared caption.
- Label directly. Put labels on the data where you can, rather than forcing readers to decode a legend.
- Mind colour. Do not rely on colour alone to distinguish groups — use shapes or patterns too, so the figure survives colour-blind readers and greyscale printing.
- Set type large enough. Axis text that is legible on your monitor is often too small once the figure is reduced to a single column.
- Start bar charts at zero. A bar shows a value by its length, so a cut-off axis exaggerates differences.
Tables done right
A good table is easy to read down a column. Keep decimal places consistent, align numbers on the decimal point, give every column a clear header with units, and define abbreviations in a footnote. If a cell has no data, say why (a dash for "not applicable" is not the same as a blank for "missing"). Resist the temptation to pack twenty columns into one table — a table that needs a magnifying glass has failed.
For categorical data, give the count and the denominator as well as the percentage — "12/48 (25%)" rather than "25%" — so a reader can check the arithmetic.
Adjusting images without crossing into manipulation
Changes to brightness and contrast are acceptable when applied to the whole image, applied equally to the controls, and when they do not obscure, eliminate or misrepresent any information in the original. Removing, adding, moving or selectively enhancing any feature is not. Altering an image that way is falsification under the usual definition of research misconduct, such as the one used by the US Office of Research Integrity (ORI).
- Keep the unprocessed original of every image, because an editor may ask for it.
- Mark any join between lanes or fields with a visible dividing line, and say so in the caption.
- State the software and the adjustments used in the Methods.
- For clinical photographs and scans, crop out identifying features rather than adding eye bars, strip the file metadata, and obtain the patient's written consent for publication.
The grey areas are covered in where image adjustment ends and manipulation begins. The identifiers hidden inside image files are covered in de-identifying clinical images and DICOM exports.
Captions, alt text and permissions
Every figure and table needs a caption that stands alone: a reader who sees only the figure should understand what it shows, including what any abbreviations and error bars mean. Add a short alt-text description too — it makes your work accessible to screen-reader users and helps search engines index it. Finally, if you are reusing a figure from elsewhere, sort out the rights before submission: an open-licence figure needs attribution, but an all-rights-reserved figure needs written permission from the copyright holder.
If a figure has error bars, the caption must say what they show — for example standard deviation, standard error, a confidence interval or an interquartile range — and give the number of observations behind each point.
Caption (invented example; the data are not real): Figure 3. Length of hospital stay by surgical approach. Points show individual patients (open, n = 22; laparoscopic, n = 25). Horizontal lines show medians; error bars show interquartile ranges.
A pre-submission checklist
- Every figure and table is cited in the text, in order.
- No figure and table present the same data.
- Resolution meets the journal's minimum at print size; graphs are vector where possible.
- Each caption stands alone and defines its abbreviations.
- Colour is not the only way groups are distinguished.
- Reuse permissions or attributions are in place.
- Every error bar is defined, and n is stated.
- Clinical images carry no identifiers, in the picture or in the file metadata.
Run that list before you submit and your paper will not just read better — it will look like the work of someone who takes their evidence seriously.
Once reviewers ask for changes, figure version control across revision rounds keeps each file traceable. To make figures readable for everyone, including colour-blind readers, see accessibility in scientific publishing.
A graphical abstract is a single image summarising the paper's main finding; how to make a graphical abstract covers what it must carry, its layout, size and file type.
Updated 17 September 2026: We expanded this guide with image-adjustment limits, error-bar and caption rules, table denominators and a longer checklist.