You design something on screen. It looks right. It goes to a printer and comes back duller, and the blue in particular has gone somewhere between navy and disappointment. Somebody tells you the problem is RGB versus CMYK, which is true, and then the explanation stops.
Every article on this topic says the same two things. Screens make color by adding light, presses make color by subtracting it with ink, and CMYK has a smaller gamut. All correct. None of it tells you what you actually want to know, which is how much smaller, which colors you lose, and whether the color you just picked is one of them.
So I built a working model of a four color offset press from the published ink standards and measured the gap. The short version:
- A press can reproduce 53 percent of sRGB closely enough that you would not see the difference. The other 47 percent is gone.
- For colors that do not make it, the median shift is 7.65 CIEDE2000, which is roughly three times the point where a person stops squinting and says those are two different colors.
- Cyans and magentas fare worst, with over 90 percent of the saturated ones unprintable. Yellows are almost untouched at 2.8 percent.
- Of everything clipping takes away, 57 percent is saturation and only 19 percent is hue. Print does not change your colors so much as tire them out.
- Around 6 percent of what a press can print is outside sRGB entirely. CMYK is not a subset of RGB, whatever the diagrams say.
The one real difference between the two models
Both systems are trying to hit the same target, which is the three signals your cone cells produce. They start from opposite ends.
RGB is additive. A screen starts black and emits red, green and blue light. Turn all three up and you get white. This is what every pixel on the device you are reading this on is doing, and it is why hex color codes are three numbers: one dial per emitter.
CMYK is subtractive. Paper starts white and reflects everything. Ink removes wavelengths. Cyan ink absorbs red light, magenta absorbs green, yellow absorbs blue. Lay down all three and in theory you absorb everything and get black. In practice you get a muddy brown around L*23, which is exactly why the K exists. Black ink is there because the chemistry does not cooperate.
This is the same additive and subtractive split that makes primary colors such a confusing topic, and it is the reason a color mixing chart for paint gives different answers than mixing light.
Here is the part that gets skipped. The gamut difference is not caused by additive versus subtractive in the abstract. It is caused by the fact that real ink is bad at its job. An ideal cyan ink would absorb all red light and no green or blue. Actual cyan ink absorbs some green, reflects some red, and sits at a lightness of 55 on a 100 point scale. Every limitation below traces back to pigment chemistry, not to the color model. If someone invented perfect inks tomorrow, the CMYK gamut would be enormous.
How I measured this
A vague claim about gamut size is easy to make and useless to act on, so the model needs to be specific about which press it describes.
I used the aim values from ISO 12647-2, the standard that specifies what four color offset printing on coated paper is supposed to look like. It defines the measured color of each solid ink and each two ink overprint. Cyan sits at L*55 a*-37 b*-50. Magenta at L*48 a*74 b*-3. Yellow at L*89 a*-5 b*93. The red you get from magenta over yellow lands at L*47 a*68 b*48, and so on for the rest.
Those nine measured points seed a Yule-Nielsen modified Neugebauer model, which is the standard way to predict what a halftone press does at partial ink coverage. The Neugebauer equations treat any printed patch as a statistical mixture of the sixteen possible ink overlaps, weighted by how likely each overlap is given the dot areas. The Yule-Nielsen exponent corrects for light scattering sideways inside the paper, which is why halftones print darker than the geometry alone predicts.
From there I sampled the CMYK space on a fine grid, capped total ink at 330 percent the way a real press would, converted every result into CIELAB under D50, and did the same for 262,144 evenly spaced sRGB colors. Then for each sRGB color I found the closest thing the press could produce and measured the distance in CIEDE2000, which is the modern color difference formula built to match what people actually perceive.
Two things worth stating plainly. The model reproduces its own seed values to 0.00 CIEDE2000, which confirms the arithmetic but is not evidence about the parts between them. And I ran the whole measurement again with the Yule-Nielsen exponent set anywhere from 1.0 to 2.5, a range that covers newsprint through glossy stock. The headline moved between 47 and 54 percent. The conclusion does not depend on tuning.
Everything is media relative, meaning paper white counts as white. If you measure in absolute terms then paper itself is only L*95 and the whole thing looks worse, but that is a complaint about paper rather than about CMYK.
Nearly half of sRGB does not survive the trip
Setting the bar at 2.3 CIEDE2000, a common working threshold for a difference an ordinary viewer notices side by side:
- 53.2 percent of sRGB prints within that threshold.
- 46.8 percent does not. For that group the median error is 7.65 and the mean is 8.10.
- The worst color in the sRGB cube misses by 18.55. At that distance you are not comparing shades, you are comparing colors that need separate names.
- Tighten the bar to 1.0, where almost nobody sees a difference, and only 43.7 percent makes it.
The tempting summary is that CMYK holds about half the colors of RGB. That is roughly right but it hides something, and the something turns out to be the most interesting result here. Hold that thought.
Which colors die, ranked
Averages across the whole cube are misleading because the damage is wildly concentrated. Restricting to genuinely saturated colors, chroma above 40, and splitting by hue:
- Cyan. 93.6 percent unprintable, median miss 8.96. Worst hue on the wheel.
- Magenta and violet. 91.2 percent unprintable, median miss 8.62, worst single case 18.55.
- Green. 69.0 percent unprintable, median miss 5.02.
- Blue. 59.9 percent unprintable, median miss 2.85.
- Red. 17.7 percent unprintable, median miss 0.64.
- Orange. 16.2 percent unprintable, median miss 0.71.
- Yellow. 2.8 percent unprintable, median miss 0.51. Essentially safe.
There is a neat irony in that list. Cyan and magenta are two of the four inks, and they are the two hues the press handles worst. It is not a contradiction. The ink defines one point, a fairly dark and only moderately saturated point, and a screen can go far brighter in the same direction. Nothing in the CMYK gamut reaches the bright saturated cyan a monitor produces by firing green and blue emitters at full power, because no amount of cyan ink on white paper is both that light and that colorful. Adding ink to raise saturation also darkens, always. That tradeoff is the whole shape of the problem.
Yellow escapes because yellow ink is unusually good. It sits at L*89, bright enough to stay near the top of the lightness range while still being strongly colored. It is the one place where the chemistry cooperates.
Some colors you have probably specified
Distances from a few familiar sRGB values to the nearest printable color:
- Pure magenta #FF00FF: 17.38. Pure green #00FF00: 16.52. Pure cyan #00FFFF: 14.65. Pure blue #0000FF: 14.07. The RGB primaries and secondaries are mostly a fantasy on paper.
- Neon lime #CCFF00: 9.96. Anything marketed as neon is a warning sign.
- Pure yellow #FFFF00: 5.36. Pure red #FF0000: 4.38. Bad, but in a different league from the blues and greens.
- Spotify green #1DB954: 3.73. Facebook blue #1877F2: 4.76. Hot pink #FF69B4: 4.58. Royal blue #4169E1: 4.68. Brand colors chosen on screens land squarely in visible-shift territory, which is why brand guidelines carry a separate print value.
- Slack aubergine #4A154B: 1.48. Mid grey #808080: 0.64. White: 0.00. Dark, desaturated and neutral colors sail through.
The practical rule falls straight out of that list. The further a color is from grey, the more likely print cannot have it. Neutrals are free.
What clipping actually takes away
Knowing a color moves by 7.65 units does not tell you which way it moved. So I decomposed every out of gamut shift into its lightness, chroma and hue components:
- Saturation: 56.7 percent of the squared difference.
- Lightness: 23.9 percent.
- Hue: 19.4 percent.
In plain terms, the median out of gamut color goes from a chroma of 72.7 to 47.7. A third of its saturation is simply deleted. Meanwhile the median hue rotates by 5.84 degrees, and even at the 90th percentile it only reaches 12.86 degrees, which on a 360 degree wheel is a nudge.
This matches what people report and rarely articulate. Nobody comes back from the printer saying their red turned green. They say it looks flat, or washed out, or lifeless. That complaint is chroma loss, and now it has a number attached. Your hue survives. Your intensity does not.
It also explains why the effect is worse for some designs than others. A layout built from muted tones is nearly immune. A layout built on one screaming accent color is entirely dependent on the color that print is least able to give you.
CMYK is not inside RGB
Search for a gamut diagram and you will find a big RGB triangle with a smaller CMYK blob tucked inside it. That picture is wrong, and it is wrong in a way that has practical consequences.
Running the comparison backwards, 6.05 percent of what the press can print falls outside sRGB altogether. Squeeze those colors into sRGB and you lose a median of 2.13 CIEDE2000, with a worst case of 9.26. Where do they live?
- Blues: 43.8 percent of the outside-sRGB set.
- Cyans: 31.6 percent.
- Greens: 19.1 percent. Yellows make up the rest.
Cyan ink and the deep blue you get from cyan over magenta are more saturated at their own lightness than anything sRGB can display. So the two gamuts overlap heavily and each pokes out of the other. sRGB owns the bright saturated region across most hues. CMYK owns a slice of deep blue green that your monitor has literally never shown you.
The consequence: a soft proof on a standard monitor is not a neutral window onto the print. Some of what the press can do gets flattened on the way to your eyes, and those are the same blues and cyans people worry about most. If you have ever thought a printed piece looked richer than the proof did, you were not imagining it.
This is also a nice reminder that a color space is a container rather than a fact about color. The same physical ink can be inside one and outside another, in the same way that two objects can match under one light and split under another, which is metamerism.
The result I did not expect
Measure the raw volume each gamut occupies in CIELAB and the press comes out at roughly 60 percent of sRGB (58 to 62 depending on how finely you grid it). That fits the standard story. CMYK is much smaller.
But CIELAB volume is not the same as usable color, because your eye is not equally sensitive everywhere in it. A just noticeable difference is a small step near grey and a much larger one out in the saturated regions. CIEDE2000 knows this: its SC and SH terms grow with chroma, which is a formal way of saying that highly saturated colors need to move further before you notice.
So I weighted the volume by local JND size, which counts distinguishable colors rather than cubic units. On that measure the press holds about 76 percent of what sRGB holds, stable at 76.0 to 76.5 across grid resolutions.
Read those two numbers together. The press gives up 40 percent of the volume but only 24 percent of the distinguishable colors. The region CMYK cannot reach is precisely the region where your eye is worst at telling one color from the next. Measuring the sRGB side on its own, I put that figure at roughly 380,000 distinguishable colors, so scaling by the ratio here a coated offset press holds somewhere near 290,000. That number moves with the threshold you pick, and how many colors you can see turns almost entirely on that choice.
That reframes the whole comparison. Print is not the impoverished cousin of the screen. It surrenders a large chunk of space that happens to be cheap in perceptual terms. The reason it still hurts is that designers do not choose colors uniformly at random from a cube. They choose the loud ones, deliberately, because loud colors do work in a layout. The gamut loses the least valuable colors by area and the most valuable ones by intent.
What to do about it
Work in RGB, convert once, at the end. The usual advice and the measurements support it. RGB is the larger container across most hues, editing in it avoids compounding rounding errors, and converting early destroys information you cannot get back.
Get your profile from the printer, not from a dropdown. Everything above describes coated offset under ISO 12647-2. Uncoated stock, newsprint and digital presses all have different and usually smaller gamuts. Newsprint in particular is far worse than these figures because the paper is dark and the ink spreads.
Test the accent color first, not the whole palette. Given how concentrated the damage is, checking one saturated cyan tells you more than checking twelve muted neutrals which were never at risk.
If a color must survive, pick it inside the gamut from the start. Choosing a bright screen color and hoping the conversion is kind means letting an algorithm redesign your brand. Better to pick something the press can hit and let the screen version be the one that gets slightly richer.
For anything critical, use a spot color. Pantone and similar systems exist because a purpose mixed ink is not bound by what four process inks can reach. It costs more per run for exactly that reason.
Do not trust your monitor uncritically. An uncalibrated screen introduces its own error, often larger than the gamut gap you are trying to evaluate. Also worth knowing that identical colors look different against different backgrounds, which is color constancy working against you.
The thing worth remembering
RGB versus CMYK is usually taught as a rule to follow: screens use one, print uses the other, do not mix them up. The measurements suggest a better framing. These are two different sets of physical limitations, one set by how brightly three emitters can glow and one set by how cleanly four pigments can absorb, and neither fully contains the other.
The gap between them is about half of sRGB, and it costs you saturation rather than hue, and it is concentrated in cyan, magenta and green while leaving yellow and every neutral untouched. That is more useful than knowing CMYK is smaller, and it is specific enough to act on before the proof comes back rather than after.
If you would like a feel for how big 7.65 CIEDE2000 really is, that is the kind of gap our color memory game asks you to close from memory, and Hex makes you work in the same three numbers a screen actually uses. Getting a sense of the scale by eye makes every number above land differently.