Making effective scientific figures comes down to a handful of core skills: choosing the right tool, following design principles that put your data front and center, and meeting the technical standards journals actually enforce. Whether you’re plotting experimental results, assembling microscopy panels, or building a schematic of a biological process, the workflow is more learnable than it looks. Here’s how to approach each part of the process.
Pick the Right Tool for the Job
No single piece of software handles every type of scientific figure well. The best choice depends on what you’re making.
For data plots and graphs, most researchers start in the environment where they already analyze data. Python (with Matplotlib or Seaborn), R (with ggplot2), and MATLAB all produce publication-quality plots and give you precise control over every visual element. These are free or already available through most institutions, and scripting your figures means they’re reproducible if your data changes.
For schematics, diagrams, and illustrated models, you need a vector graphics editor. Adobe Illustrator is the industry standard, but Inkscape is a free alternative that handles the same file formats. Vector graphics scale to any size without losing sharpness, which matters when journals resize your figure to fit a column. BioRender has become popular for biological pathway diagrams because it offers a library of pre-drawn cellular components, though it requires a subscription for publication use.
For microscopy and image analysis, ImageJ (and its extended version, Fiji) remains the go-to open-source platform. It handles everything from cell counting to 3D reconstruction through a massive plugin ecosystem. CellProfiler is built specifically for high-throughput cell image analysis, while 3D Slicer handles volumetric data like CT or MRI scans when your work moves into three dimensions.
Most polished figures combine tools. A typical workflow might be: generate the plot in R, export it as a vector file (SVG or PDF), then arrange multiple panels and add labels in Illustrator or Inkscape.
Design Around the Data, Not Around Decoration
The most influential design principle in scientific visualization comes from Edward Tufte’s concept of the “data-ink ratio.” The idea is straightforward: every drop of ink in your figure should represent data. Anything that doesn’t, what Tufte called “chartjunk,” should be removed. That means eliminating unnecessary gridlines, background shading, 3D effects on bar charts, and redundant labels. A clean figure communicates faster and more honestly than a busy one.
Tufte himself noted that chartjunk should be removed “within reason.” You still need axis labels, legends, and enough visual structure for the reader to orient themselves. The goal isn’t minimalism for its own sake. It’s making sure nothing competes with the actual data for the reader’s attention. If an element doesn’t help the reader interpret the result, cut it.
Axis Scaling Can Distort Your Story
One of the easiest ways to accidentally mislead a reader is through axis choices. A set of values plotted from 0 to 100 on the y-axis will look like they’re all roughly the same. The same values plotted from 80 to 100 will look dramatically different. Even when the axis is properly labeled, the visual impression of the bars or lines dominates how people interpret the figure.
Start your axes at zero for bar charts. For line graphs where the trend matters more than the absolute values, a truncated axis can be appropriate, but label it clearly. Use logarithmic scales when your data spans several orders of magnitude, such as gene expression levels or bacterial growth curves, since a linear scale would compress the smaller values into invisibility. Always label the axis to indicate a log transformation so readers aren’t comparing magnitudes they think are linear.
Resolution and File Format Requirements
Journals reject figures that look fine on screen but fall apart in print. The standard minimum resolution for submission is 300 DPI (dots per inch). That applies to photographs and continuous-tone images like microscopy. Line art, such as graphs and diagrams with sharp edges and text, typically needs 600 DPI or higher to avoid jagged lines. Combination figures (photos with text overlaid) should meet the higher threshold.
The simplest way to avoid resolution problems is to work in vector formats (SVG, EPS, PDF) for any figure that isn’t a photograph. Vector files are resolution-independent, meaning they stay sharp at any size. For photographs and microscopy images, export as TIFF at 300 DPI or above. JPEG compression introduces artifacts, particularly around sharp edges and text, so avoid it for final submissions unless the journal specifically requests it.
Before submitting, run a grayscale visibility test. Many journals still print in black and white, and color distinctions that are obvious on screen can vanish entirely in grayscale. Convert your figure to grayscale temporarily and check whether you can still tell every element apart.
Typography That Survives Reduction
Journal figures are almost always shrunk to fit a one- or two-column layout. Text that was legible at full size on your monitor can become unreadable in the published version. Science magazine’s guidelines specify a 6-point minimum font size after the figure is reduced to its final printed dimensions, with a minimum line width of 0.5 points. Those are good baselines for any journal.
Use a sans-serif font for all labels, axis titles, and legends. Helvetica and Arial are the most widely accepted. Reserve serif fonts like Times New Roman for the manuscript body text, not the figures. Sans-serif letters are easier to read at small sizes and against complex backgrounds. Keep your labeling consistent across all panels: same font, same size hierarchy, same capitalization style.
Choose Colors That Everyone Can See
Roughly 8% of men and 0.5% of women have some form of color vision deficiency, most commonly difficulty distinguishing red from green. If your figure relies on a red-versus-green color scheme, a significant portion of your audience will struggle to read it.
For continuous color maps, the Viridis palette (available in R, Python, and MATLAB) is designed to be perceptually uniform and colorblind-safe. It progresses from dark purple through blue and green to yellow, and it remains distinguishable in grayscale. For categorical comparisons where you need distinct colors, the Okabe-Ito palette is widely recommended. If you must use contrasting warm and cool tones, blue versus orange or blue versus red are safer alternatives to red versus green.
Beyond palette choice, reinforce color with a second visual channel. Use different line styles (solid, dashed, dotted), marker shapes (circles, triangles, squares), or pattern fills in bar charts. That way, even if two colors look identical to a colorblind reader, the shapes or patterns still communicate the distinction.
Microscopy Figures Need Scale Bars
For any microscopy or imaging figure, always include a scale bar directly on the image. A scale bar is more reliable than stating magnification in the caption, because magnification changes depending on the display size and the reader’s screen or print resolution. Place the bar where it’s clearly visible against the image background: a white bar on a dark region, or a black bar on a light region. Include the dimension (e.g., “50 μm”) on or immediately next to the bar, and make sure both the bar and text remain legible after the figure is reduced to its final size.
When your data involves multiple scales, use insets. A zoomed-out view gives context, while an inset at higher magnification shows the detail that supports your claim. Each view needs its own scale bar.
Image Processing: What’s Allowed and What Isn’t
Adjusting brightness and contrast is acceptable, as long as you apply the change uniformly across the entire image and don’t obscure, eliminate, or misrepresent any information. Think of it like adjusting exposure in a darkroom: you’re making the image easier to see, not changing what’s in it. A useful technique is a contrast stretch (sometimes called a histogram stretch), where the darkest pixel in the image is remapped to black and the brightest to white. This spreads the tonal range without adding or removing information.
Avoid auto-contrast, auto-levels, and auto-color tools in Photoshop. These tend to over-process images, and it’s nearly impossible to report exactly what they changed, which creates problems during peer review. If you make manual adjustments, note them in the figure legend or methods section.
Cropping is generally acceptable for centering a region of interest or trimming empty space. It crosses into manipulation when it removes context that would change the interpretation, like cropping out dead cells to make a treatment look more effective, or removing gel bands that contradict your hypothesis. The guiding question is simple: does removing this information change what a viewer would conclude from the figure?
AI-Generated Figures and Journal Policies
Generative AI tools can now produce scientific illustrations from text prompts, but major publishers have drawn firm lines. Nature’s current policy prohibits AI-generated images in published papers, citing unresolved copyright and research integrity concerns. The narrow exceptions require that any AI-generated content be clearly labeled within the image itself and reviewed on a case-by-case basis.
Non-generative AI tools, such as machine learning algorithms that enhance resolution, segment structures, or denoise microscopy images, fall under different rules. Their use should be disclosed in the figure caption so reviewers can assess whether the processing introduced artifacts. The distinction matters: using AI to sharpen a real image is treated differently from using AI to create an image from scratch.
If you’re using any AI tool in your figure pipeline, check the specific journal’s policy before submission. These policies are evolving quickly and vary between publishers.
Assembling Multi-Panel Figures
Most published figures contain multiple panels, labeled A through D (or more), arranged in a logical reading order: left to right, top to bottom. Each panel should be understandable on its own but contribute to a single coherent narrative when viewed together.
Label panels with bold uppercase letters in the top-left corner. Use the same font across all panels. Align edges precisely, both horizontally and vertically, so the figure looks intentional rather than cobbled together. Leave enough white space between panels that they don’t bleed into each other, but not so much that the figure feels disconnected. Most vector editors have alignment and distribution tools that make this straightforward.
Write figure legends that can stand alone. A reader should be able to understand what the figure shows without reading the full paper. State what’s being measured, what the axes represent, what each color or symbol means, and what the sample size is. Keep it factual and compact, but don’t leave gaps that force the reader to hunt through the methods section.