Scientific Figure Rules for Journal Submission

Every major journal treats figures as a submission requirement in their own right, checked separately from manuscript text before a paper is even sent to peer review. A figure that fails a technical check — wrong resolution, an unsupported file format, a missing panel label — can trigger an editorial-office bounce-back before a reviewer ever sees the work. A figure that has been edited in a way that misrepresents the underlying data raises a different and more serious problem: a research-integrity concern that can follow a paper long after publication. This guide covers both halves of “figure rules” — the technical specification (resolution, file format, numbering, panel labeling, color accessibility) and the integrity boundary (what counts as acceptable image processing versus manipulation that publishers, the Committee on Publication Ethics (COPE), and the International Committee of Medical Journal Editors (ICMJE) treat as misconduct) — and shows, with real examples, how the specifics differ by journal and publisher even though the underlying principles are consistent.

Why figure preparation is a compliance issue, not just a design one

Figure requirements exist for two separate reasons that are easy to conflate. The first is production: journals print and display figures at fixed physical sizes, across print, PDF, and web, and an image that looks fine on a laptop screen can be unusable at a journal’s actual column width if it wasn’t prepared at sufficient resolution. The second is scientific record-keeping: a figure is data, not illustration, and any processing applied to it has to leave the underlying result honestly represented. ICMJE’s Recommendations on scientific misconduct explicitly name “data falsification including deceptive manipulation of images” as a form of misconduct, on the same footing as fabricating numerical data. That is why a guide to “figure rules” has to cover DPI and file formats in the same breath as image-integrity policy — a technically perfect figure that has been selectively edited is still a problem, and a rule violation on either side can stall or unwind a publication.

Resolution and DPI: what journals are actually asking for

Dots per inch (DPI) describes how many pixels are packed into each printed inch — it is not a property of a digital file on its own, only of a file at a specified physical output size. Increasing the DPI value in image-editing software without increasing the underlying pixel count (upsampling) does not add real detail; it just spreads the same information over more reported dots, and journals’ production staff can generally tell. The only reliable way to meet a resolution requirement is to capture or export the original figure at the required pixel dimensions for its intended print size.

Requirements are not identical across figure types, and most journals separate photographic/halftone images from line art (graphs, diagrams, schematics) because the two degrade differently at low resolution — photographs with continuous tonal gradients can look acceptable at lower DPI than line art, where fine strokes and small text alias into jagged edges well before 300 dpi. Three real, currently published examples:

Nature (Springer Nature): figures should be provided “in RGB color and at 300 dpi or higher resolution,” per Nature’s own guide for initial and revised submissions.
PLOS (PLOS ONE and PLOS’s other journals): figures must be submitted at 300–600 dpi — below 300 dpi results in blurred, jagged, or pixelated published figures, and resolution above 600 dpi may cause automated resizing during production.
Wiley: line art is expected at 600 dpi and photographic images/scans at 300 dpi, with a combined acceptable range of 300–600 dpi across figure types at final acceptance; a lower-resolution figure is tolerated only at initial submission, provided it is “legible to the editor and peer reviewers.”

The consistent floor across virtually every reputable journal is 300 dpi for photographic/halftone images, with line art, vector diagrams, and combination figures generally expected at a higher resolution — commonly in the 600–1200 dpi range depending on the journal. Because that range varies by publisher (and sometimes by journal within the same publisher’s family), always confirm the exact figure against the specific journal’s current author guidelines before final submission rather than assuming one publisher’s number applies elsewhere.

File formats: raster versus vector, and what each publisher accepts

Figures fall into two broad categories that most journals treat differently:

Raster (bitmap) images — pixel-based, used for photographs, micrographs, and scanned images. Common accepted formats: TIFF, PSD, high-quality PNG or JPEG. Raster images lose quality if resized upward after the fact, which is why the source resolution matters so much.
Vector images — defined by mathematical paths rather than pixels, used for graphs, charts, schematics, and line art. Common accepted formats: EPS, PDF, AI (Adobe Illustrator), SVG. Vector files are resolution-independent — they scale cleanly to any output size — which is why most journals prefer vector formats for line art wherever the figure was actually generated as a vector graphic (e.g., a plot exported from statistical or plotting software) rather than rasterized first.

What this looks like at three real publishers:

PLOS accepts TIFF or EPS only for final figure files — no other format — and notes that TIFF is generally easier for authors to produce correctly than EPS.
Nature accepts vector formats (.ai, .eps, .pdf, .ps, .svg) for fully editable line art, layered .psd or .tif for editable layered artwork, and .psd, .tif, .png, or .jpg for flattened bitmap images.
Wiley prefers PDF for line art and PNG or TIFF for photographic images, but accepts “any standard file type” as a fallback, with the instruction “when in doubt, submit a PDF.”

A figure saved as a screenshot, or exported from a slide deck at on-screen resolution, will fail almost every journal’s technical check regardless of format — the format alone does not fix a resolution or file-size problem.

Figure numbering and in-text callout conventions

Two conventions are close to universal across journals, even where the specific formatting details vary:

Sequential numbering by first citation. Figures are numbered in the order they are first mentioned in the manuscript text, not the order they might appear logically in a results narrative if that differs. Skipping a number, or citing Figure 3 in the text before Figure 2, is one of the most common reasons a manuscript is returned during editorial pre-checks.
File names must match in-text citations and captions exactly. PLOS, for example, specifies that a file named “Fig1.tif” must correspond to the in-text citation “Fig 1” and the caption label “Fig 1.” — a mismatch between the file name, the callout in the body text, and the caption is treated as an error even if the content is otherwise correct.

Callout style itself is not standardized: some journals abbreviate to “Fig.” in running text and spell out “Figure” only at the start of a sentence or in the caption; others use “Figure” throughout. Panel-level callouts follow the same inconsistency — “Fig. 1a,” “Fig. 1(a),” and “Figure 1A” are all used by different journals for the same thing (a reference to panel A within Figure 1). Match whatever convention the target journal’s own guidelines specify rather than defaulting to a habit from a previous submission.

Panel labeling conventions

Multi-part figures (Figure 1 containing panels A, B, C…) need consistent, unambiguous labeling because panels are referenced individually both in the caption and in the body text. Widely shared conventions:

Each panel gets a single capital or lowercase letter — PLOS’s own author guidance specifies a “lettered panel label: for example, (A) or (a)” — used consistently for that letter’s case throughout the whole manuscript.
Labels are placed in a consistent position across all panels (typically the upper-left corner), in a consistent font and size, and must not be placed over or obscure any data, image content, or background — PLOS states labeling and annotation must be performed “without obscuring any data or background.”
The caption text should describe what each lettered panel shows, in the same letter order used in the figure itself.

Where a figure combines panel types that were prepared separately (e.g., a photomicrograph panel next to a bar-chart panel), the resolution requirement that applies is generally whichever is higher for that figure type — a combination figure is commonly held to the line-art standard even where one panel is itself a photograph, because the lowest-resolution element in a combined file sets the visible quality ceiling for the whole figure.

Colorblind-accessible color choices

Red-green color vision deficiency affects roughly 1 in 12 men and a much smaller share of women, commonly cited around 1 in 200 — a large enough fraction of any journal’s readership that color choice in figures is treated by several major journals as a technical requirement, not a stylistic preference. Two concrete, widely adopted tools:

The Okabe-Ito palette — an eight-color, colorblind-distinguishable palette originally proposed by Masataka Okabe and Kei Ito, popularized across the scientific-publishing world after Bang Wong’s 2011 Nature Methods “Points of view” column on color blindness (Wong, B. Nat. Methods 8, 441 (2011)). It is now commonly used as the default categorical palette in plotting libraries specifically because it remains distinguishable under the most common forms of color vision deficiency, unlike a default red/green/blue categorical scheme.
Perceptually uniform continuous colormaps (e.g., viridis) — for heatmaps, contour plots, or any figure encoding a continuous variable in color, a perceptually uniform, colorblind-safe colormap avoids both the accessibility problem and the distortion risk of a “rainbow”/jet colormap, where equal steps in data value do not correspond to equal perceived steps in color.

Nature’s own author guidance is explicit on this point: it advises against “the use of red and green for contrast,” recommends “green and magenta or other accessible color palettes” instead, and separately states that “use of the rainbow color scale should be avoided.” Treat this as representative of where major-journal policy is heading generally, not a Nature-specific quirk — check the target journal’s own guidelines, since not every journal states an explicit accessibility rule even where reviewers may still flag a red/green figure informally.

Avoiding image manipulation issues — where figure prep becomes a research-integrity question

The line between acceptable image processing and problematic manipulation is not about whether an image was adjusted at all — nearly every figure is. It is about whether the adjustment is applied uniformly, is disclosed, and leaves the image an honest representation of the underlying data. The framework most publishers still use traces back to a 2004 Journal of Cell Biology editorial by Mike Rossner and Kenneth Yamada (“What’s in a picture? The temptation of image manipulation,” J Cell Biol 166(1):11–15, 2004), which set out the standard that adjustments to brightness, contrast, or color balance are acceptable only if applied to every pixel in the image equally, and only if they do not obscure, eliminate, or misrepresent any information present in the original — including the background. That principle, in close to its original wording, has been carried into the editorial policies of many journals since.

Practical implications for authors preparing figures:

Global adjustments only. Brightness, contrast, and color-balance changes must be applied to the whole image, not selectively to one region (e.g., brightening a faint band on a gel without brightening the rest of the lane equally).
Non-linear adjustments must be disclosed. Changes to gamma settings or other non-linear processing need to be stated, typically in the figure legend or methods section — silent non-linear adjustment is treated as a red flag even when the underlying intent was benign.
Splicing must be marked. If a figure combines lanes or images from different parts of the same gel, or from different exposures, that has to be visually indicated (e.g., with a dividing line) and disclosed in the caption — not presented as a single continuous image.
Keep the originals. Editors and integrity investigators routinely ask for unprocessed source files. PLOS, for example, requires authors submitting on or after 1 April 2026 to provide the original, uncropped, and minimally adjusted images supporting every blot and gel result shown in a figure or supporting file — a direct response to how often published figures are later found to have been selectively cropped.

When a concern is raised after publication, COPE’s own “Inappropriate image manipulation in a published article” flowchart (version 2, published 20 May 2024) is what many editors work from to decide the response — whether the manipulation affected the paper’s conclusions, whether a correction is sufficient, or whether the case warrants an expression of concern or retraction while an institutional investigation proceeds. See CASRAI’s guides on the image manipulation and image duplication dictionary entries, COPE flowcharts explained, and COPE Guidelines explained for how this fits into the broader research-integrity and retraction process. The takeaway for authors preparing figures is upstream of any of that: process every figure the same way you would want a reviewer, an editor, or a future investigator to be able to reconstruct — uniformly, disclosed, and against a retained original.

How figure requirements vary by journal and publisher

The underlying principles above (resolution floors by figure type, vector-vs-raster format choice, consistent panel labeling, colorblind-accessible palettes, uniform and disclosed processing) are close to universal. The specific numbers and accepted formats are not, and checking a journal’s own current author guidelines before final submission is not optional — the table below shows how much three real, major publishers differ on figures alone.

Requirement	Nature (Springer Nature)	PLOS	Wiley
Minimum resolution (photos/halftone)	300 dpi or higher, RGB	300–600 dpi	300 dpi (final)
Minimum resolution (line art)	Vector strongly preferred over a raster minimum	300–600 dpi (same range, TIFF/EPS only)	600 dpi (final)
Accepted formats	.ai/.eps/.pdf/.ps/.svg (vector); .psd/.tif/.png/.jpg (raster)	TIFF or EPS only	PDF preferred (line art); PNG/TIFF preferred (photos); “any standard type” accepted as fallback
Color/accessibility guidance	Avoid red/green contrast; use green/magenta or other accessible palettes; avoid rainbow scale	RGB (8-bit/channel) or grayscale only	Not separately specified in general guidelines
Panel labeling	Consistent typeface across all figures (Arial/Helvetica)	Lettered panel label, e.g. (A) or (a); must not obscure data	Arabic numerals for multi-part callouts; labels legible at final size
File size / dimension limits	Standard widths 90 mm (single column) / 180 mm (double column)	≤10 MB per file; width 789–2,250 px at 300 dpi	≤10 MB per figure; ≤500 MB total submission; 80–180 mm width

Note that “Nature” and “PLOS” are each families of journals with a shared production standard rather than a single title — the same general figures policy typically applies across a publisher’s portfolio, but always verify against the specific journal (not just the publisher) before final submission, since sub-journals occasionally diverge on specifics like file-size caps.

A practical pre-submission checklist

Confirm the target journal’s current figure guidelines directly — publisher-level defaults (as summarized above) are a starting point, not a substitute for the specific journal’s author instructions.
Export photographic/halftone figures at the required native resolution (do not upsample) and line art as vector files wherever the source software allows it.
Use the accepted file format(s) exactly — TIFF/EPS-only journals will reject a PNG or PDF regardless of resolution.
Number figures in the order first cited in the text; make sure file names, in-text callouts, and caption labels match exactly.
Label multi-panel figures with a consistent letter/number scheme, positioned consistently, never obscuring data.
Choose a colorblind-accessible palette (e.g., Okabe-Ito for categorical data, viridis for continuous data) instead of a default red/green or rainbow scheme.
Apply any brightness/contrast/color adjustments uniformly across the whole image, disclose non-linear adjustments in the legend, mark any splicing, and retain unprocessed originals in case an editor requests them.

Frequently asked questions

What DPI do most journals require for figures?

Most major journals set a floor of 300 dpi for photographic or halftone images, and a higher standard — commonly in the 600–1200 dpi range, or a strong preference for resolution-independent vector formats — for line art, graphs, and diagrams. The exact number varies by publisher: Nature asks for 300 dpi or higher, PLOS specifies a 300–600 dpi range for all figure types, and Wiley separates the two (300 dpi for photos, 600 dpi for line art). Always confirm the current number against the specific journal’s own author guidelines.

What file format should scientific figures be submitted in?

It depends on the journal and the figure type. Vector formats (EPS, PDF, AI, SVG) are generally preferred for line art and graphs because they scale without quality loss; raster formats (TIFF, PSD, high-resolution PNG or JPEG) are used for photographs and scanned images. Some journals, like PLOS, restrict final figure files to TIFF or EPS only — no other format is accepted at that stage.

How should I label panels in a multi-part figure?

Use a single, consistent letter per panel (uppercase or lowercase — PLOS specifies “(A) or (a)” as acceptable formats), positioned in the same location on every panel, in the same font and size, and never placed over image content or data. The figure caption should describe each panel in the same letter order used in the image.

What colors should I avoid for colorblind accessibility?

Avoid relying on red and green together to distinguish categories — this is the combination most affected by the most common forms of color vision deficiency. Several major journals, including Nature, explicitly recommend green/magenta or other colorblind-safe combinations instead, and advise against rainbow or “jet” colormaps for continuous data because they are both hard to interpret and inaccessible to colorblind readers. The Okabe-Ito palette (categorical data) and viridis (continuous data) are widely used, verified alternatives.

What counts as inappropriate image manipulation?

Selectively adjusting brightness, contrast, or color in only part of an image (rather than uniformly across the whole image); failing to disclose non-linear adjustments such as gamma changes; splicing together lanes or exposures without indicating the join; and any adjustment that obscures, eliminates, or misrepresents information present in the original image. ICMJE treats deceptive image manipulation as a form of data falsification, and COPE maintains a dedicated editorial flowchart for handling suspected cases after publication.

Do all journals have the same figure requirements?

No. The underlying principles (resolution floors, format categories, panel-labeling logic, accessible color choices, and image-integrity standards) are broadly shared across reputable journals, but the specific numbers, accepted file formats, and size limits differ by publisher and sometimes by individual journal within the same publisher’s portfolio. Nature, PLOS, and Wiley — three of the largest scholarly publishers — each specify different exact DPI ranges and accepted formats, as detailed above. Treat any general figure-preparation guide, including this one, as a starting checklist to verify against the specific target journal’s current author instructions.

Why do figures matter for research integrity, not just formatting?

Because a figure is a representation of research data, not decoration. ICMJE explicitly names deceptive image manipulation as a category of scientific misconduct alongside data fabrication and falsification, and COPE maintains editorial guidance specifically for responding to image-manipulation concerns raised about published articles. Getting the technical formatting right (resolution, format, labeling) is a production requirement; keeping the processing honest and disclosed is a research-integrity requirement — both are checked, separately, by most journals.