Default Chart Colors and Color Blindness: 9 Charting Tools Tested
Under simulated red-green color blindness, 7 of the 45 color pairs in Matplotlib's default palette fall below a ΔE2000 of 10, the threshold our palette checker flags as a collision. The worst pair is the second and third series colors: under protanopia, orange and green go from ΔE 55 to under 2.
That is the protanopia simulation. Under deuteranopia, the other red-green type, the same two colors are ΔE 14.5 apart, which is not below the threshold. In the independent check, DaltonLens' implementation of the same Machado 2009 model and two different models, Viénot 1999 and Brettel 1997, each count 7 as well.
How it was measured: both colors simulated as complete dichromacy (Machado 2009), the difference taken with CIEDE2000, each pair scored by the worse of protanopia and deuteranopia. Engine readings for this pair: ΔE 55.2 with typical color vision, 1.4 under protanopia. Hex values read from Matplotlib's source on November 18, 2026.
Matplotlib is one case. We ran the default series colors of 9 charting tools through the same test: Matplotlib, seaborn, Microsoft Office (Excel, PowerPoint, Word), Tableau, Power BI, ggplot2, Plotly (plotly.py), Chart.js and R (base graphics). That is 12 palettes. For each one we simulated protanopia and deuteranopia, measured every pair of colors with CIEDE2000, and counted the pairs that fall below 10, the threshold our palette collision tool uses for a collision. Every one of the 12 default palettes has at least one pair below 10.
Running one tool's defaults through a simulator has been done before, usually ending in a screenshot and an adjective. What this page adds is a number for every pair, one method across tools, and a link on every row that recomputes the result, so a figure you quote can be checked.
The figures simulate complete dichromacy, which is the most severe case, and the thresholds are working values for design review. This is not a clinical finding and nobody's vision was tested. See Method and limits.
Results at a glance
- Matplotlib (tab10): 7 of 45 pairs below 10, the first with 3 series. D3's category10 is the same 10 colors.
- Excel and PowerPoint, 2013–2022 theme: 1 of 15 (orange and green, series 2 and 6). The 2023 theme: 2 of 15.
- Tableau 10: 10 of 45. Under deuteranopia, red and green (series 3 and 5) go from ΔE 62 to under 1.
- ggplot2: with 3 classes, 1 of 3; with 8 classes, 8 of 28.
- Power BI (first 10 data colors): 5 of 45. Plotly: 6 of 45.
- Controls: Okabe-Ito 0 of 28; seaborn's "colorblind" palette 6 of 45.
How to read the numbers
- ΔE2000 (CIEDE2000) measures how different two colors look. Bigger is easier to tell apart.
- Below 10 counts as a collision. That is the cut-off our tool uses; in thin lines and small markers, colors that close are easy to mix up. Between 10 and 20 counts as risky: distinguishable side by side, unreliable when the marks are small or far apart.
- Each pair is simulated twice, for protanopia (red-blind) and deuteranopia (green-blind), and scored by the lower of the two results.
- First collision at series is the smallest number of series for which the tool's first colors already include a pair below 10. For a tool that hands out a fixed list in order, a chart with fewer series than that has no collision on this test. ggplot2 is the exception: its colors change with the number of classes, so its rows have to be read one class count at a time.
Summary: 12 default palettes
| Palette | Colors, pairs | Below 10collisions | 10 to 20risky | First collision at series | Closest pairΔE: typical → simulated | DaltonLens check |
|---|---|---|---|---|---|---|
| Matplotlibtab10 (default axes.prop_cycle) D3 category10: same values, same orderOpen in checker | 1045 pairs | 7 | 8 | 3 | orange / greenseries 2 and 3, protanopia55.2 → 1.4 | 7 / 7 / 7 |
| seaborn (deep)deep (default of set_theme() and of seaborn.objects) Open in checker | 1045 pairs | 8 | 11 | 3 | blue / purpleseries 1 and 5, protanopia13.4 → 2.0 | 8 / 8 / 8 |
| Office 2013–2022 themeOffice 2013 – 2022 theme, accents 1–6 Open in checker | 615 pairs | 1 | 4 | 6 | orange / greenseries 2 and 6, deuteranopia47.2 → 6.2 | 1 / 1 / 1 |
| Office theme, 2023 onwardOffice theme (2023 onward), accents 1–6 Open in checker | 615 pairs | 2 | 2 | 5 | orange / greenseries 2 and 6, deuteranopia54.6 → 5.1 | 2 / 2 / 2 |
| Office 2007–2010 themeOffice 2007 – 2010 theme, accents 1–6 Open in checker | 615 pairs | 2 | 3 | 4 | olive green / orangeseries 3 and 6, deuteranopia36.6 → 4.9 | 2 / 2 / 2 |
| Tableau 10Tableau 10 (Automatic categorical palette) Open in checker | 1045 pairs | 10 | 11 | 5 | red / greenseries 3 and 5, deuteranopia62.4 → 0.7 | 10 / 10 / 10 |
| Power BIBase theme data colors, first 10 Open in checker | 1045 pairs | 5 | 11 | 4 | red / greenseries 8 and 10, deuteranopia73.4 → 5.1 | 5 / 5 / 5 |
| ggplot2, 3 classesscale_colour_discrete(), 3 classes Open in checker | 33 pairs | 1 | 0 | 2 | salmon / greenseries 1 and 2, deuteranopia70.0 → 5.9 | 1 / 1 / 1 |
| ggplot2, 8 classesscale_colour_discrete(), 8 classes Open in checker | 828 pairs | 8 | 4 | 3 | cyan / pinkseries 5 and 8, deuteranopia56.0 → 0.9 | 9 / 6 / 7 |
| Plotly (plotly.py)“plotly” template colorway Open in checker | 1045 pairs | 6 | 13 | 4 | blue / purpleseries 1 and 4, protanopia12.6 → 0.7 | 6 / 7 / 7 |
| Chart.jsColors plugin palette Open in checker | 721 pairs | 3 | 4 | 4 | teal / grayseries 5 and 7, protanopia23.6 → 5.7 | 3 / 3 / 3 |
| R palette()palette() default (“R4”) Open in checker | 828 pairs | 2 | 3 | 6 | green / yellowseries 3 and 7, protanopia30.9 → 4.3 | 2 / 2 / 2 |
ΔE values are CIEDE2000 to one decimal: the difference with typical color vision, then the lower of the protanopia and deuteranopia simulations. The last column, "DaltonLens check", is an independent check: the number of pairs below 10 according to three simulations from DaltonLens. In order, they are a second implementation of the Machado 2009 model our engine uses, then two different models, Viénot 1999 and Brettel 1997. ΔE2000 for all three comes from colour-science. "Open in checker" loads the same colors into the tool, which lists every pair.
Have a palette of your own?
Paste the hex codes and get the same pair-by-pair table, with a link you can share.
Check your paletteMatplotlib
These 10 colors are what Matplotlib hands to the first 10 lines when you set nothing.
Palette: tab10 (default axes.prop_cycle)
#1f77b4, #ff7f0e, #2ca02c, #d62728, #9467bd, #8c564b, #e377c2, #7f7f7f, #bcbd22, #17becf
7 of 45 pairs fall below 10 under simulated red-green color blindness, and 8 more sit between 10 and 20. The first collision appears once a chart has 3 series.
- Orange and green (series 2 and 3)ΔE 55.2 → 1.4protanopia; nearly identical after simulation
- Blue and purple (series 1 and 5)ΔE 26.4 → 1.8protanopia; below 10 under both types; nearly identical after simulation
- Orange and olive (series 2 and 9)ΔE 35.9 → 3.4deuteranopia
- Pink and cyan (series 7 and 10)ΔE 53.8 → 4.1deuteranopia
- Green and red (series 3 and 4)ΔE 71.8 → 4.8deuteranopia
- Blue and pink (series 1 and 7)ΔE 42.8 → 9.4protanopia; close to the threshold
- Red and brown (series 4 and 6)ΔE 16.2 → 9.7protanopia; close to the threshold
Note: Matplotlib's source carries two comments about where these come from: “# These colors are from Tableau” above TABLEAU_COLORS (the same ten values under their tab: names, in lib/matplotlib/_color_data.py) and “# Categorical palettes from Vega:” above _tab10_data (in lib/matplotlib/_cm.py). They are the ten colors of the palette Tableau called Tableau 10 before version 10.0, which Tableau's help now calls Tableau Classic 10.
Version checked and where the values were read
Default since Matplotlib 2.0; checked at v3.11.2
_tab10_data in lib/matplotlib/_cm.py at tag v3.11.2: floats, converted to hex here and matching the hex in each line's comment. The default is declared in lib/matplotlib/rcsetup.py (_Param("axes.prop_cycle", default=cycler("color", [...])), the same ten triples); the sample matplotlibrc documents it as cycler(color='tab10'), a name colors.py maps to _cm._tab10_data. The v2.0.0 and v3.9.0 rc files list the same ten values inline.
Primary source · values read November 18, 2026 · Open in checker
D3: D3's schemeCategory10, from d3-scale-chromatic. Same ten values in the same order as Matplotlib's tab10; Plotly.js uses the same ten when no template is applied. Primary source
A 2024 Towards Data Science article plotted the first 8 of these colors, simulated full green-blindness on the rendered image with DaltonLens' Brettel 1997, and described what the simulated image looked like. Its overall finding was that “the colors are generally still distinguishable, although they appear closer together,” with two examples: “blue and purple (C0 and C4) are barely distinguishable. The same holds for red, green, and brown (C2, C3, C5).” Our deuteranopia figures for the pairs it names: blue and purple (ΔE 6.1) and green and red (ΔE 4.8) are below 10; green and brown (ΔE 16.6) and red and brown (ΔE 14.8) are between 10 and 20, the band this page calls risky. The lowest-scoring pair on this page, orange and green, is a protanopia result.
seaborn
seaborn has its own default, deep: 10 colors with different hex values from Matplotlib's. The note below says when it applies.
Palette: deep (default of set_theme() and of seaborn.objects)
#4c72b0, #dd8452, #55a868, #c44e52, #8172b3, #937860, #da8bc3, #8c8c8c, #ccb974, #64b5cd
8 of 45 pairs fall below 10 under simulated red-green color blindness, and 11 more sit between 10 and 20. The first collision appears once a chart has 3 series.
- Blue and purple (series 1 and 5)ΔE 13.4 → 2.0protanopia; below 10 under both types; nearly identical after simulation
- Red and brown (series 4 and 6)ΔE 21.8 → 4.5deuteranopia; below 10 under both types
- Orange and green (series 2 and 3)ΔE 46.3 → 4.8protanopia; below 10 under both types
- Pink and sky blue (series 7 and 10)ΔE 41.2 → 5.2deuteranopia; below 10 under both types
- Green and brown (series 3 and 6)ΔE 31.0 → 7.7deuteranopia
- Orange and khaki (series 2 and 9)ΔE 23.9 → 7.7deuteranopia
- Green and khaki (series 3 and 9)ΔE 24.5 → 8.1protanopia
- Green and red (series 3 and 4)ΔE 61.5 → 8.1deuteranopia
Note: It depends on the interface. With the function interface (sns.lineplot and the like) deep applies only after sns.set_theme(); without that call seaborn draws with Matplotlib's current color cycle, which by default is Matplotlib's own ten colors. The seaborn.objects interface (so.Plot) uses deep with no call at all, because its default theme sets the color cycle to deep. In both, a plot with more categories than the cycle has colors switches to seaborn's husl palette.
Version checked and where the values were read
Checked at seaborn v0.13.2
SEABORN_PALETTES["deep"] in seaborn/palettes.py at tag v0.13.2. Two defaults point at it: set_theme(..., palette="deep") in seaborn/rcmod.py, and the seaborn.objects theme in seaborn/_core/plot.py, whose defaults include "axes.prop_cycle": cycler("color", color_palette("deep")).
Primary source · values read November 18, 2026 · Open in checker
Excel and PowerPoint (Microsoft Office theme colors)
Charts in Excel, PowerPoint and Word take their series colors from the document theme: accent 1 for the first series through accent 6 for the sixth. A document keeps the theme it was created with, so more than one theme is in circulation, and each is listed separately below.
Office 2013–2022 theme
Palette: Office 2013 – 2022 theme, accents 1–6
#4472c4, #ed7d31, #a5a5a5, #ffc000, #5b9bd5, #70ad47
1 of 15 pairs falls below 10 under simulated red-green color blindness, and 4 more sit between 10 and 20. The first collision appears once a chart has 6 series.
- Orange and green (series 2 and 6)ΔE 47.2 → 6.2deuteranopia; below 10 under both types
Note: Files with this theme all have these six colors, but in two orders. Files saved by Office 2013 (AppVersion 15) have accent 1 = #5B9BD5 and accent 5 = #4472C4, so a chart in them starts with the lighter blue; so do 14 of the 20 files with these colors saved by version 16 that were checked. The order recorded here, the one Microsoft ships today, appears only in files saved by version 16, and which build introduced it was not pinned down. Pairwise results are the same either way; anything that depends on series order is not. From the seventh series on, Office reuses the six with lightness variations.
Version checked and where the values were read
Shipped today as “Office 2013 - 2022 Theme”, the default until the 2023 theme; checked in Excel for Mac 16.115. Older files have accents 1 and 5 the other way around (see note)
a:accent1 to a:accent6 in theme/theme/theme1.xml inside “Office 2013 - 2022 Theme.thmx”, shipped in Excel for Mac 16.115 under Contents/Resources/Office Themes (sha256 bc3d3bf6…). The linked workbook, written by Excel 16.0300 and published by Microsoft in dotnet/Open-XML-SDK, has the same six values in xl/theme/theme1.xml. Series order: Office charts carry cs:colorStyle meth="cycle" id="10" listing accent1 to accent6 (charts/colors1.xml in Microsoft's test files; [MS-ODRAWXML] 2.8.3.2).
Primary source · values read November 18, 2026 · Open in checker
Office theme, 2023 onward
Palette: Office theme (2023 onward), accents 1–6
#156082, #e97132, #196b24, #0f9ed5, #a02b93, #4ea72e
2 of 15 pairs fall below 10 under simulated red-green color blindness, and 2 more sit between 10 and 20. The first collision appears once a chart has 5 series.
- Orange and green (series 2 and 6)ΔE 54.6 → 5.1deuteranopia; below 10 under both types
- Dark blue and magenta (series 1 and 5)ΔE 37.5 → 5.5deuteranopia; below 10 under both types
Note: Documents keep the theme they were created with, so the 2013–2022 colors and these are both in circulation.
Version checked and where the values were read
The new default Office theme, announced in July 2023 for Beta Channel builds (Version 2307 on Windows, 16.75 on Mac); checked in Excel for Mac 16.115
a:accent1 to a:accent6 in theme/theme/theme1.xml inside “Office Theme.thmx”, shipped in Excel for Mac 16.115 under Contents/Resources/Office Themes (sha256 2e5736d6…). The linked Flat OPC document, written by Word 16 and published by Microsoft in dotnet/Open-XML-SDK, has the same six values in its /word/theme/theme1.xml part.
Primary source · values read November 18, 2026 · Open in checker
Office 2007–2010 theme
Palette: Office 2007 – 2010 theme, accents 1–6
#4f81bd, #c0504d, #9bbb59, #8064a2, #4bacc6, #f79646
2 of 15 pairs fall below 10 under simulated red-green color blindness, and 3 more sit between 10 and 20. The first collision appears once a chart has 4 series.
- Olive green and orange (series 3 and 6)ΔE 36.6 → 4.9deuteranopia; below 10 under both types
- Blue and purple (series 1 and 4)ΔE 20.5 → 5.7deuteranopia; below 10 under both types
Note: The default Office theme in files saved by Office 2007 and 2010. Microsoft still ships it as the “Office 2007-2010” color scheme.
Version checked and where the values were read
Office 2007 and 2010 (AppVersion 12 and 14); read from the color scheme still shipped in Excel for Mac 16.115
a:accent1 to a:accent6 in Theme Colors/Office2007-2010.xml, shipped in Excel for Mac 16.115 under Contents/Resources/Office Themes (sha256 39b573d4…). Same values as Accent1Color to Accent6Color of the “Office Theme” built at lines 1599–1604 of the linked Microsoft sample, and as the themes of Microsoft's test documents saved by Office 2007 and 2010 in dotnet/Open-XML-SDK.
Primary source · values read November 18, 2026 · Open in checker
Tableau
Tableau 10 is the palette Tableau applies automatically to a categorical field. A chart with two series uses only its first two colors and a chart with 10 uses all of them. Those are different questions, so the result is given in two layers.
With only the first two colors, blue and orange, the pair's lower result is ΔE 48.4 (protanopia): clear of both thresholds. With three series, the count below 10 is 0 and the count between 10 and 20 is 1; the lowest is orange and red at ΔE 14.1 (deuteranopia). The first pair below 10 arrives when series 5 is added.
With all 10 colors in use, 10 of the 45 pairs in Tableau 10 are below 10, and red and green (series 3 and 5) go from ΔE 62 to under 1 under deuteranopia.
Palette: Tableau 10 (Automatic categorical palette)
#4e79a7, #f28e2b, #e15759, #76b7b2, #59a14f, #edc948, #b07aa1, #ff9da7, #9c755f, #bab0ac
10 of 45 pairs fall below 10 under simulated red-green color blindness, and 11 more sit between 10 and 20. The first collision appears once a chart has 5 series.
- Red and green (series 3 and 5)ΔE 62.4 → 0.7deuteranopia; nearly identical after simulation
- Pink and gray (series 8 and 10)ΔE 20.6 → 1.3protanopia; below 10 under both types; nearly identical after simulation
- Red and brown (series 3 and 9)ΔE 18.8 → 2.1protanopia; below 10 under both types; nearly identical after simulation
- Teal and gray (series 4 and 10)ΔE 24.5 → 3.2protanopia; below 10 under both types
- Teal and pink (series 4 and 8)ΔE 46.0 → 3.9protanopia
- Orange and green (series 2 and 5)ΔE 44.6 → 6.2protanopia
- Blue and purple (series 1 and 7)ΔE 29.1 → 6.6protanopia
- Green and brown (series 5 and 9)ΔE 35.4 → 7.5deuteranopia
- Orange and yellow (series 2 and 6)ΔE 22.1 → 8.4deuteranopia
- Teal and purple (series 4 and 7)ΔE 38.4 → 8.6deuteranopia
Note: d3's schemeTableau10 is not identical: it has #F28E2C, #EDC949, #AF7AA1 and #BAB0AB where Tableau Desktop writes #F28E2B, #EDC948, #B07AA1 and #BAB0AC. Before version 10.0 the palette called Tableau 10 was the Matplotlib/D3 set; Tableau's help now calls that one Tableau Classic 10.
Version checked and where the values were read
Tableau's help says the palette was updated in version 10.0; values read from files written by Tableau Desktop 2019.1.10 and by build main.26.0802.1921
The color encoding of field [Metric] in the linked Tableau bookmark (header build main.26.0802.1921, published by Tableau in tableau/tableau-mcp): ten members on the default palette, ten <map to> values. Sorted by member name they come out blue, orange, red, teal, green, yellow, purple, pink, brown, gray, which is the order the PALETTE comment in Tableau's gen_workbook.py (tableau/tableau-ai-utilities) calls “the order Tableau hands out colors”. A workbook saved by Tableau 2019.1.10 (tableau/community-tableau-server-insights) holds the same ten values.
Primary source · values read November 18, 2026 · Open in checker
On series order: the 10 colors were read from a file written by Tableau Desktop, where they are attached to the 10 members of one field. Sorted by member name they come out blue, orange, red, teal, green, yellow, purple, pink, brown, gray, which matches the order a comment in one of Tableau's own scripts gives for how colors are handed out. We did not confirm that order in a running copy of Tableau Desktop. The count of 10 in 45 does not depend on it; the "first collision" figure does.
Power BI
Power BI takes series colors from the report theme's data colors, in order. What is measured here is the first 10 of the base theme; the note below says why that many.
Palette: Base theme data colors, first 10
#118dff, #12239e, #e66c37, #6b007b, #e044a7, #744ec2, #d9b300, #d64550, #197278, #1aab40
5 of 45 pairs fall below 10 under simulated red-green color blindness, and 11 more sit between 10 and 20. The first collision appears once a chart has 4 series.
- Red and green (series 8 and 10)ΔE 73.4 → 5.1deuteranopia
- Dark blue and dark purple (series 2 and 4)ΔE 15.2 → 5.4protanopia; below 10 under both types
- Orange and green (series 3 and 10)ΔE 60.1 → 5.7deuteranopia
- Gold and green (series 7 and 10)ΔE 32.8 → 8.2protanopia
- Pink and purple (series 5 and 6)ΔE 23.3 → 8.2protanopia
Note: The base theme lists 41 data colors (32 in Fluent 2, equal to the first 32 of the 41); the first ten are recorded here. Why ten: the example theme on Microsoft Learn's “Create custom report themes in Power BI Desktop” lists these same ten values as its dataColors, and ten is the size of the Matplotlib, Plotly and Tableau palettes these are set beside.
Version checked and where the values were read
The same first ten in every base theme checked, from CY19SU06 to Fluent2-CY26SU11: eight themes, four of them found only in third-party repositories
The first ten entries of dataColors (41 in all) in StaticResources/SharedResources/BaseThemes/CY24SU10.json, the base theme Power BI writes into a report, taken from the blank report Microsoft ships in microsoft/fabric-cli. The CY19SU06, CY23SU04 and Fluent 2 base themes begin with the same ten. Microsoft Learn (“Use report themes in Power BI”) states that series take the theme's data colors in order.
Primary source · values read November 18, 2026 · Open in checker
ggplot2
ggplot2 spaces the hues evenly around the HCL color wheel, starting at 15°, at chroma 100 and luminance 65, so every color except the first changes with the number of classes.
So each row below is a different set of colors, one row for each class count from 2 to 8. ggplot2's first color, #f8766d, is often treated as its red. It is called salmon here because that is closer to how the swatch looks.
| Classes | Colors, pairs | Below 10collisions | 10 to 20risky | First collision at series | Closest pairΔE: typical → simulated | DaltonLens check |
|---|---|---|---|---|---|---|
| 2 classes Open in checker | 21 pair | 0 | 0 | none | salmon / cyanseries 1 and 2, protanopia55.5 → 25.3 | 0 / 0 / 0 |
| 3 classes Open in checker | 33 pairs | 1 | 0 | 2 | salmon / greenseries 1 and 2, deuteranopia70.0 → 5.9 | 1 / 1 / 1 |
| 4 classes Open in checker | 46 pairs | 1 | 1 | 2 | salmon / yellow-greenseries 1 and 2, deuteranopia58.6 → 9.3 | 1 / 0 / 1 |
| 5 classes Open in checker | 510 pairs | 3 | 1 | 2 | blue / magentaseries 4 and 5, deuteranopia47.3 → 2.7 | 3 / 2 / 2 |
| 6 classes Open in checker | 615 pairs | 4 | 2 | 3 | ochre / greenseries 2 and 3, protanopia28.2 → 3.9 | 4 / 4 / 4 |
| 7 classes Open in checker | 721 pairs | 5 | 4 | 3 | ochre / greenseries 2 and 3, deuteranopia28.4 → 3.1 | 5 / 5 / 5 |
| 8 classes Open in checker | 828 pairs | 8 | 4 | 3 | cyan / pinkseries 5 and 8, deuteranopia56.0 → 0.9 | 9 / 6 / 7 |
A plot with fewer classes gets a different set of colors from ggplot2, not the first few of a larger set. So "first collision at series" in this table is a position inside that row's own set: in the 8-class set, for example, the first 3 colors already contain a pair below 10.
With 2 classes, salmon and cyan are ΔE 25.3 apart under protanopia, the lower of the two results: clear of both thresholds. With 3, 1 of 3 pairs is below 10; the closest are salmon and green (series 1 and 2), which go from ΔE 70.0 to 5.9 under deuteranopia. With 8, 8 of 28 pairs are below 10; the closest are cyan and pink (series 5 and 8), which go from ΔE 56.0 to 0.9 under deuteranopia.
ggplot2, 3 classes
Palette: scale_colour_discrete(), 3 classes
#f8766d, #00ba38, #619cff
1 of 3 pairs falls below 10 under simulated red-green color blindness, and 0 more sit between 10 and 20.
- Salmon and green (series 1 and 2)ΔE 70.0 → 5.9deuteranopia
Version checked and where the values were read
ggplot2 4.0.4 with scales 1.4.0
Output of scales::pal_hue()(3) run in R 4.6.0 (WebR 0.6.1, scales 1.4.0, farver 2.1.2). pal_hue in R/pal-hue.R at tag v1.4.0 defaults to h = c(0, 360) + 15, c = 100, l = 65, and ggplot2 v4.0.4 R/scale-colour.R gives scale_colour_discrete the fallback palette pal_hue(). Base R grDevices::hcl() over the same hues returns identical strings; ggplot2's own test snapshots show the same values for 3 and 6 classes.
Primary source · values read November 18, 2026 · Open in checker
ggplot2, 8 classes
Palette: scale_colour_discrete(), 8 classes
#f8766d, #cd9600, #7cae00, #00be67, #00bfc4, #00a9ff, #c77cff, #ff61cc
8 of 28 pairs fall below 10 under simulated red-green color blindness, and 4 more sit between 10 and 20.
- Cyan and pink (series 5 and 8)ΔE 56.0 → 0.9deuteranopia; nearly identical after simulation
- Blue and purple (series 6 and 7)ΔE 35.8 → 2.2deuteranopia; below 10 under both types; nearly identical after simulation
- Ochre and yellow-green (series 2 and 3)ΔE 26.4 → 2.6deuteranopia; below 10 under both types; nearly identical after simulation
- Salmon and green (series 1 and 4)ΔE 68.5 → 3.4deuteranopia
- Purple and pink (series 7 and 8)ΔE 14.7 → 7.4protanopia
- Yellow-green and green (series 3 and 4)ΔE 16.5 → 8.7protanopia
- Salmon and yellow-green (series 1 and 3)ΔE 58.6 → 9.3deuteranopia; close to the threshold; at or above 10 in at least one of the DaltonLens simulations
- Ochre and green (series 2 and 4)ΔE 38.6 → 9.8protanopia; close to the threshold; at or above 10 in at least one of the DaltonLens simulations
Version checked and where the values were read
ggplot2 4.0.4 with scales 1.4.0
Output of scales::pal_hue()(8) run in R 4.6.0 (WebR 0.6.1, scales 1.4.0, farver 2.1.2). pal_hue in R/pal-hue.R at tag v1.4.0 defaults to h = c(0, 360) + 15, c = 100, l = 65, and ggplot2 v4.0.4 R/scale-colour.R gives scale_colour_discrete the fallback palette pal_hue(). Base R grDevices::hcl() over the same hues returns identical strings; ggplot2's own test snapshots show the same values for 3 and 6 classes.
Primary source · values read November 18, 2026 · Open in checker
Plotly
These are the colors of plotly.py's default template, which Plotly Express also uses. As with Tableau, the result is given in two layers.
With only the first two colors, blue and red, the pair's lower result is ΔE 53.5 (protanopia): clear of both thresholds. With three series, the count below 10 is 0 and the count between 10 and 20 is 1; the lowest is red and green at ΔE 18.5 (deuteranopia). The first pair below 10 arrives when series 4 is added.
The closest pair is blue and purple (series 1 and 4), which go from ΔE 12.6 to 0.7 under protanopia. The two are close to begin with: ΔE 12.6 with typical color vision, already under 20.
Palette: “plotly” template colorway
#636efa, #ef553b, #00cc96, #ab63fa, #ffa15a, #19d3f3, #ff6692, #b6e880, #ff97ff, #fecb52
6 of 45 pairs fall below 10 under simulated red-green color blindness, and 13 more sit between 10 and 20. The first collision appears once a chart has 4 series.
- Blue and purple (series 1 and 4)ΔE 12.6 → 0.7protanopia; below 10 under both types; nearly identical after simulation
- Cyan and light magenta (series 6 and 9)ΔE 50.0 → 2.2deuteranopia; below 10 under both types; nearly identical after simulation
- Green and pink (series 3 and 7)ΔE 76.9 → 2.8deuteranopia; nearly identical after simulation
- Light green and yellow (series 8 and 10)ΔE 24.9 → 6.0protanopia; below 10 under both types
- Orange and yellow (series 5 and 10)ΔE 18.2 → 6.6deuteranopia; below 10 under both types
- Orange and light green (series 5 and 8)ΔE 40.9 → 6.7deuteranopia
Note: This is plotly.py, including Plotly Express, which takes its default colors from the template's colorway. Plotly.js on its own, with no template, falls back to colorAttrs.defaults in src/components/color/attributes.js (checked at v4.3.0): the same ten values as D3's category10 and Matplotlib's tab10.
Version checked and where the values were read
Default template since plotly.py 4.0; checked at v7.1.0
layout.colorway in plotly/package_data/templates/plotly.json at tag v7.1.0; plotly/io/__init__.py sets templates._default = "plotly". The same ten values are Plotly in _plotly_utils/colors/qualitative.py, and the v4.0.0 template has the same colorway.
Primary source · values read November 18, 2026 · Open in checker
Chart.js
Chart.js takes its default colors from its Colors plugin, 7 in all. What is measured here is each color at full strength; the note below says which chart types use it that way and which use it at 50% opacity.
Palette: Colors plugin palette
#36a2eb, #ff6384, #ff9f40, #ffcd56, #4bc0c0, #9966ff, #c9cbcf
3 of 21 pairs fall below 10 under simulated red-green color blindness, and 4 more sit between 10 and 20. The first collision appears once a chart has 4 series.
- Teal and gray (series 5 and 7)ΔE 23.6 → 5.7protanopia
- Orange and yellow (series 3 and 4)ΔE 16.3 → 6.6deuteranopia; below 10 under both types
- Blue and purple (series 1 and 6)ΔE 28.4 → 7.2deuteranopia
Note: Applied when the Colors plugin is registered (it always is in the UMD build and chart.js/auto) and no colors are set on the datasets, in the chart options or in Chart.defaults. Doughnut and pie charts color each slice at full strength and polar area charts at 50% opacity; every other chart type gives each dataset one of these as its border color and the same color at 50% opacity as its fill. Without the plugin every dataset is the same translucent gray, rgba(0, 0, 0, 0.1).
Version checked and where the values were read
Colors plugin, in Chart.js since v4.0.1 (not in v4.0.0); checked at v4.6.1
BORDER_COLORS in src/plugins/plugin.colors.ts at tag v4.6.1, written as rgb() and converted to hex here.
Primary source · values read November 18, 2026 · Open in checker
R (base graphics)
This is the default of palette() since R 4.0.0, the colors base graphics uses for col = 1, 2, 3 and so on.
Palette: palette() default (“R4”)
#000000, #df536b, #61d04f, #2297e6, #28e2e5, #cd0bbc, #f5c710, #9e9e9e
2 of 28 pairs fall below 10 under simulated red-green color blindness, and 3 more sit between 10 and 20. The first collision appears once a chart has 6 series.
- Green and yellow (series 3 and 7)ΔE 30.9 → 4.3protanopia
- Blue and magenta (series 4 and 6)ΔE 47.7 → 7.4deuteranopia
Note: The first entry is black, so a series drawn with col = 1 is black; the six chromatic colors are entries 2 to 7 and the eighth is a gray. R's NEWS for 4.0.0 describes the set as “less saturated” with “better accessibility properties”.
Version checked and where the values were read
Default since R 4.0.0; checked in R 4.6.0
DefaultPalette[8] in src/library/grDevices/src/colors.c at tag R-4-6-0, stored as 0xAABBGGRR and converted here. palette() run in R 4.6.0 prints black, #DF536B, #61D04F, #2297E6, #28E2E5, #CD0BBC, #F5C710, gray62; black and gray62 resolve to #000000 and #9E9E9E.
Primary source · values read November 18, 2026 · Open in checker
Not in the table: Google Sheets
Google Sheets is not in the table. We could not find its default series colors in anything Google itself publishes, and this page only uses values that were read from a tool's own source, read from a file the tool wrote, or returned by the tool's own function (the case for ggplot2). It never uses a list copied from somewhere else.
Controls: two palettes built for color blindness
For comparison, here is the same measurement on two palettes meant to hold up under color vision deficiency. Neither is the default of any tool in the table above.
| Palette | Colors, pairs | Below 10collisions | 10 to 20risky | First collision at series | Closest pairΔE: typical → simulated | DaltonLens check |
|---|---|---|---|---|---|---|
| seaborn colorblindcolorblind Open in checker | 1045 pairs | 6 | 10 | 4 | light pink / sky blueseries 7 and 10, protanopia43.0 → 5.7 | 6 / 5 / 6 |
| Okabe-ItoMasataka Okabe (Jikei Medical School) & Kei Ito (University of Tokyo), 2002, revised 2008 Open in checker | 828 pairs | 0 | 7 | none | orange / yellowseries 2 and 5, deuteranopia21.7 → 11.5 | not run |
seaborn's "colorblind" palette is not free of collisions on this test: 6 of its 45 pairs are below 10. 4 of them score within one unit of the threshold (9.0 to 9.6); the other 2 score 5.7 (protanopia) and 6.1 (protanopia). A pair's score is its lower ΔE across the protanopia and deuteranopia simulations. The three DaltonLens simulations count 6, 5 and 6.
Okabe-Ito, all 8 colors including black, has 0 of 28 pairs below 10 and 7 between 10 and 20; the closest are orange and yellow at ΔE 11.5 under deuteranopia. It was not part of the DaltonLens check, which is why that column says "not run" for it.
seaborn colorblind
Palette: colorblind
#0173b2, #de8f05, #029e73, #d55e00, #cc78bc, #ca9161, #fbafe4, #949494, #ece133, #56b4e9
6 of 45 pairs fall below 10 under simulated red-green color blindness, and 10 more sit between 10 and 20. The first collision appears once a chart has 4 series.
- Light pink and sky blue (series 7 and 10)ΔE 43.0 → 5.7protanopia
- Green and tan (series 3 and 6)ΔE 40.5 → 6.1protanopia
- Orange and vermilion (series 2 and 4)ΔE 17.0 → 9.0deuteranopia; close to the threshold
- Pink and sky blue (series 5 and 10)ΔE 43.1 → 9.1deuteranopia; close to the threshold
- Green and gray (series 3 and 8)ΔE 24.9 → 9.2deuteranopia; close to the threshold; at or above 10 in at least one of the DaltonLens simulations
- Blue and pink (series 1 and 5)ΔE 40.1 → 9.6protanopia; close to the threshold
Note: A control, not a default: seaborn's colorblind palette, selected with sns.set_palette('colorblind'). The other control is Okabe-Ito, one of the reference palettes in this site's palette checker.
Version checked and where the values were read
Checked at seaborn v0.13.2
SEABORN_PALETTES["colorblind"] in seaborn/palettes.py at tag v0.13.2.
Primary source · values read November 18, 2026 · Open in checker
Okabe-Ito
Palette: Okabe-Ito
#000000, #e69f00, #56b4e9, #009e73, #f0e442, #0072b2, #d55e00, #cc79a7
0 of 28 pairs fall below 10 under simulated red-green color blindness, and 7 more sit between 10 and 20.
Method and limits
For every palette we took the colors in the order the tool assigns them to series and compared every pair. Both colors of a pair are first converted to what the Machado 2009 model predicts for protanopia and for deuteranopia at full severity; the difference between the two simulated colors is then measured with CIEDE2000 (ΔE2000) in CIE Lab. A pair's score is the lower of its two results.
A pair below 10 is counted as a collision and a pair below 20 as risky. Those two cut-offs are engineering working values for design review. They are not clinical limits, and no standard defines them.
What these numbers are not
- Not a test on people. Nobody with a color vision deficiency was asked to read these charts. We are not vision scientists or eye-care professionals; this is arithmetic on a published model.
- A worst case. The simulation is of complete dichromacy. The most common red-green type, deuteranomaly, is one the National Eye Institute describes as mild (our color blindness statistics give the proportions). For the milder types the model predicts larger differences; this page did not compute them.
- Flat patches, not charts. ΔE compares two areas of solid color. Thin lines, small markers, anti-aliasing, an uncalibrated screen or a projector all make a real chart harder to read than the number suggests; direct labels make it easier.
- Red-green only. Blue-yellow deficiency is less common and is left out of the counts here; the tool behind each row reports it separately.
The independent check
The engine's results were re-run with code we did not write. The simulation came from DaltonLens 0.1.5, three ways: its own implementation of Machado 2009, the model our engine uses, and two different models, Viénot 1999 and Brettel 1997. ΔE2000 came from colour-science 0.4.7. The first is a check on our implementation; the other two show whether a result depends on the choice of model. The last column of the summary, ggplot2 and controls tables gives the three counts in that order.
- Simulated colors: the engine's Machado 2009 output differs from DaltonLens' implementation of the same model by at most 1 per 8-bit channel.
- ΔE2000: across 2,465 comparisons the engine differs from scikit-image 0.26.0 by at most 0.016 and from colour-science by at most 0.33. 10 comparisons differ from colour-science by more than 0.1; every one involves a neutral gray swatch, and the smallest ΔE at which it happens is 19.5. For values under 10, the range that decides a collision, the largest difference is 0.002. These comparison counts include D3 category10, which repeats Matplotlib's colors, so those comparisons are counted twice.
- Leaving out D3 category10, the check covers 18 palettes. Counting one result per palette, pair and type, 87 are below 10 in the engine and in all three DaltonLens simulations, and 76 of those are under 9 in all four.
- Pairs that sit close to 10 can land on either side of it when the model changes, and even with the same model when the last rounding step differs. So for these palettes the DaltonLens counts are not all equal to the engine's: Plotly (plotly.py) (engine 6, DaltonLens 6 / 7 / 7); ggplot2, 4 classes (engine 1, DaltonLens 1 / 0 / 1); ggplot2, 5 classes (engine 3, DaltonLens 3 / 2 / 2); ggplot2, 8 classes (engine 8, DaltonLens 9 / 6 / 7); seaborn colorblind (engine 6, DaltonLens 6 / 5 / 6).
Where each palette comes from
The hex values were read from the tool's own source at a pinned version, or from a file the tool itself wrote, on November 18, 2026. ggplot2 is the exception: its colors are not stored in any file, and are what its own function, scales::pal_hue(), returned when run in R. Each palette with its own block above has a fold-out naming the exact file and variable. Two kinds of palette have no block: ggplot2's other class counts, which were produced the same way as the ones shown, and D3, whose source is linked in the Matplotlib section.
| Palette | Colors | Primary source | Values read |
|---|---|---|---|
| Matplotlibtab10 (default axes.prop_cycle) | 10 | github.com | 2026-11-18 |
| Office 2013–2022 themeOffice 2013 – 2022 theme, accents 1–6 | 6 | github.com | 2026-11-18 |
| Office theme, 2023 onwardOffice theme (2023 onward), accents 1–6 | 6 | github.com | 2026-11-18 |
| Office 2007–2010 themeOffice 2007 – 2010 theme, accents 1–6 | 6 | github.com | 2026-11-18 |
| Plotly (plotly.py)“plotly” template colorway | 10 | github.com | 2026-11-18 |
| ggplot2, 2 classesscale_colour_discrete(), 2 classes | 2 | github.com | 2026-11-18 |
| ggplot2, 3 classesscale_colour_discrete(), 3 classes | 3 | github.com | 2026-11-18 |
| ggplot2, 4 classesscale_colour_discrete(), 4 classes | 4 | github.com | 2026-11-18 |
| ggplot2, 5 classesscale_colour_discrete(), 5 classes | 5 | github.com | 2026-11-18 |
| ggplot2, 6 classesscale_colour_discrete(), 6 classes | 6 | github.com | 2026-11-18 |
| ggplot2, 7 classesscale_colour_discrete(), 7 classes | 7 | github.com | 2026-11-18 |
| ggplot2, 8 classesscale_colour_discrete(), 8 classes | 8 | github.com | 2026-11-18 |
| Tableau 10Tableau 10 (Automatic categorical palette) | 10 | github.com | 2026-11-18 |
| Power BIBase theme data colors, first 10 | 10 | github.com | 2026-11-18 |
| seaborn (deep)deep (default of set_theme() and of seaborn.objects) | 10 | github.com | 2026-11-18 |
| Chart.jsColors plugin palette | 7 | github.com | 2026-11-18 |
| R palette()palette() default (“R4”) | 8 | svn.r-project.org | 2026-11-18 |
| D3 category10schemeCategory10 | 10 | github.com | 2026-11-18 |
| seaborn colorblindcolorblind | 10 | github.com | 2026-11-18 |
| Okabe-ItoMasataka Okabe (Jikei Medical School) & Kei Ito (University of Tokyo), 2002, revised 2008 | 8 | jfly.uni-koeln.de | — |
Known gaps
- Tableau's series order is reconstructed, not confirmed in the running product; see the Tableau section.
- For the Office 2007–2010 theme, the order in which a chart hands out the accents is inferred. The part of a chart file that records that order exists only in files written by Office 2013 and later.
- What exactly was measured for each tool (the two accent orders found in files with the Office 2013–2022 theme, how many Power BI colors, full-strength or half-transparent Chart.js colors, plotly.py against Plotly.js, R's first color being black) is in the note under each palette.
- Color names in the text (orange, green and so on) are descriptive labels added for readability. What was measured is the hex value.
Reproducing a number
“Open in checker” on any row loads that palette into the palette collision tool, which lists the ΔE of every pair. The engine's numbers on this page (counts, ΔE values, first-collision positions) are produced by the same code when the page is built, not typed in; change a hex value or a threshold and they change with it. The DaltonLens check column and the figures under “The independent check” are different: they are read from the stored output of the cross-check run and are not recomputed when the page is built. If the engine counts recorded in that output stop matching what the engine computes now, the build fails.
What to do about it
None of this makes the default palettes unusable. It means color cannot be the only thing separating two series.
- Do not rely on color alone. Label lines directly at their ends, and give series different marker shapes and line styles. A series is then identifiable even when two colors look the same.
- Change the palette when color has to carry the distinction. On this page's test, the counts below 10 are: Okabe-Ito 0 of 28, Paul Tol's bright set 0 of 21, Paul Tol's muted set 0 of 36. All three can be loaded with one click in the palette collision tool.
- Use fewer colors. n colors make n(n−1)/2 pairs, so every added series adds pairs that all have to stay apart. The “first collision at series” column in the summary is the series count at which each default palette first contains a colliding pair, so a chart with one fewer series has none (ggplot2 aside, where each class count is its own set of colors).
- Test the palette you actually ship. Paste its hex codes into the checker, or, if the chart is already drawn, run the image through the color blindness simulator to see the whole figure under each type.
For what the two red-green types see, read the pages on protanopia and deuteranopia.
Frequently Asked Questions
Sources
- Machado, Oliveira & Fernandes (2009) — "A Physiologically-based Model for Simulation of Color Vision Deficiency," IEEE Transactions on Visualization and Computer Graphics — the model the engine uses to simulate protanopia and deuteranopia
- Sharma, Wu & Dalal (2005) — "The CIEDE2000 Color-Difference Formula: Implementation Notes, Supplementary Test Data, and Mathematical Observations," Color Research & Application — the formulation of ΔE2000 the engine implements
- DaltonLens-Python — the three simulations used for the independent check: a second implementation of the Machado 2009 model, and two different models, Viénot 1999 and Brettel 1997
- colour-science and scikit-image — the two independent ΔE2000 implementations
- Okabe & Ito — Color Universal Design — source of the Okabe-Ito control palette
- National Eye Institute (NIH) — Types of Color Vision Deficiency — deuteranomaly, the most common red-green type, is mild; blue-yellow deficiency is the less common type
- Towards Data Science, a 2024 article on making graphs accessible to colorblind readers — the earlier observation about Matplotlib's defaults, compared in the Matplotlib section
- The primary source of every palette (the tool's own code, a file it wrote, or for ggplot2 its own function) is listed in the table under Where each palette comes from.
Check the Palette You Actually Use
Paste any list of hex codes to get the same pair-by-pair numbers, or open any row above in the tool and swap colors until the collisions are gone.