Two socks in a drawer look identical. Under the bathroom light one of them is clearly browner. A sleeve matches the body of the jacket on the shop floor and stops matching on the pavement outside. A tooth crown blends in perfectly in the dentist's chair and glows slightly wrong in a photograph.
None of those objects changed. The light did, and the objects disagreed about it. That disagreement has a name.
Metamerism is when two objects with different spectral reflectance curves produce the same color response in your eye, which means they match under one light source and can stop matching under another. The two objects are called a metameric pair, or metamers.
The textbook version stops there. I wanted to know how big the effect actually gets, so I built the spectral math and measured it. A pair of reflectance curves that match perfectly under daylight, meaning a color difference of 0.0000, comes apart by a median of 9.1 CIEDE2000 under an incandescent bulb, with a worst case near 37. Roughly 89 percent of those pairs cross the threshold where a person would say the colors are obviously different. Metamerism is not a rounding error. It is one of the largest color effects there is.
Why metamerism has to exist
The reason is arithmetic, and it is more brutal than most explanations let on.
Light arriving at your eye is a spectrum. Sample it every 5 nanometres from 380 to 780, the range the standard colorimetric observer is defined over, and you have 81 numbers. That is the full physical description of what is hitting your retina.
Your eye does not receive 81 numbers. It has three cone types, and each one integrates the whole spectrum down to a single response. Eighty one numbers in, three numbers out. Color vision is a compression step with a ratio of 27 to 1, and it happens before any processing your brain does.
Once you write that as a matrix, the consequence is unavoidable. Mapping 81 dimensions onto 3 leaves a null space of 78 dimensions. Every one of those 78 directions is a change you can make to a spectrum that your eye is mathematically incapable of registering. Color scientists call a spectrum living in that null space a metameric black, because adding it to any other spectrum changes the physics and changes nothing you can see.
So metamerism is not a defect in dye chemistry or a failure of quality control. It is a direct consequence of having three cone types and a world made of continuous spectra. Any three-channel eye in any universe would have the same problem.
This is also the reason the same physical color can be described by three numbers at all, which is what makes hex color codes and every other three-value notation possible. The compression that creates metamerism is the compression that makes color practical.
How far apart a matched pair can drift
The 78 dimensions of metameric black are the theory. The practical question is how much damage they can do, so I set it up as a measurement.
The method: build the standard 1931 observer functions and the D65 daylight spectrum, restrict reflectance curves to smooth shapes that a real pigment could plausibly have, then for a given starting color solve for the pair of curves that is an exact match under D65 and separates as far as possible under CIE illuminant A. Illuminant A is the 2856 K Planckian radiator that stands in for a tungsten bulb, which is to say ordinary warm indoor light. Every result below is scored in CIEDE2000, with the observer adapted to whichever light is on, so the numbers describe a real viewer in a real room rather than a raw sensor reading.
The spread across 390 starting colors:
- Under D65, every pair scores 0.0000. That is the constraint, not a finding.
- Under illuminant A, the median separation is 9.13 CIEDE2000.
- The worst pair reaches 37.18, which is not a subtle shift. That is the distance between colors you would give different names.
- 89 percent of the pairs land above 2.3, the figure usually quoted as the point where a difference becomes obvious.
Hold the reflectance curves to something even smoother, six components rather than ten, and the effect shrinks but does not go away: median 3.15 and 62 percent still above threshold. Smooth pigments are safer than spiky ones. They are not safe.
Setting 9.13 against the discrimination threshold is the part worth sitting with. In the piece on the just noticeable difference I measured what 2.3 CIEDE2000 actually buys you: about a 3.5 point move in brightness or a 5.5 degree turn in hue. A median metameric failure of 9.13 is roughly four times that. Changing the bulb in a room can move a matched pair four just noticeable differences apart, and the objects never touched each other.
The three kinds worth knowing
Illuminant metamerism
The one above, and the one people actually run into. Same observer, same two objects, different light source. The paint shop has fluorescent overheads, your hallway has a warm LED, and the sun does something different again at 8am than it does at 4pm. This is why any decent paint counter tells you to take the sample home.
Observer metamerism
Same objects, same light, different eye. Cone pigments vary from person to person, macular pigment density varies, and lenses yellow with age. A pair that matches for you can fail for the person next to you.
There is a version of this you can measure without needing two people, because the standard observer comes in two field sizes. The 1931 functions describe a patch about 2 degrees wide, roughly a thumbnail at arm's length. The 1964 functions describe a 10 degree patch, about the size of your palm held out. Same person, different amount of retina.
Running the same optimisation against that difference: pairs that match exactly for the 2 degree observer separate by a median of 5.31 CIEDE2000 for the 10 degree observer, with a worst case of 18.11. That has a blunt practical reading. A match you verified on two small chips can fail once the same two paints are on a wall, purely because the patch got bigger on your retina.
If you want the extreme version of observer variation, the article on tetrachromacy covers what happens when someone has a fourth cone type. Four channels instead of three shrinks the null space, which means colors that are metamers for everyone else are not metamers for them.
Geometric metamerism
Same objects, same light, same eye, different angle. Anything with gloss, metallic flake or a textured weave reflects differently depending on where you stand. Car paint is the notorious case, which is why body shops check a repair from several angles rather than one.
A note on the word itself, since search results for it are a mess. Metamerism in chemistry means a type of structural isomerism, and metamerism in biology means a body built from repeated segments, as in earthworms. Three unrelated fields, one Greek root, no connection between them beyond "made of parts".
Your screen is a metamerism machine
Here is the part that matters if you spend any time matching colors on a display.
A real leaf reflects a broad, continuous spectrum. Your monitor cannot reproduce that spectrum and does not try. It has three primaries, each a fairly narrow band, and it mixes them until the three cone responses come out right. Every image you have ever looked at on a screen is a metameric forgery of the thing it depicts. It matches for the standard observer by construction, and that is the only guarantee on offer.
I modelled that too: take a smooth object reflectance, solve for the display primary mix that reproduces its color exactly for the 2 degree observer, then ask the 10 degree observer what it thinks. The 2 degree difference is 0.0000 every time, as designed. The 10 degree observer disagrees, and how much depends sharply on hue:
- Deep red, 0.33. Leaf green, 0.45. Yellow, 0.77. All invisible.
- Orange, 1.31. Violet, 1.07. Still under threshold.
- Teal, 6.42. Sky blue, 7.07. Both comfortably visible.
The failure is concentrated in blue and blue green, and the reason is that the short wavelength end is where the 2 degree and 10 degree observers disagree most, while display blue primaries are the narrowest band a monitor emits. Narrow primaries plus observer disagreement is exactly the recipe for a metamer that breaks. It is a small extra reason the blue green region is awkward, on top of the naming problem I went through in shades of teal.
The same hex on two different panels
There is a cruder version of the same problem that has nothing to do with spectra and everything to do with software, and it is worth quantifying because it affects anyone comparing color across devices.
An 8-bit triple like #FF1900 is not a color. It is an instruction, and it only becomes a color once you know which primaries the display is using. Send an sRGB value to a Display P3 panel without color management and the panel obeys with its own, wider primaries. Running that comparison over 200,000 random colors:
- Median difference 2.98 CIEDE2000, with 73 percent of colors above the visible threshold.
- Saturated colors only, median 3.61.
- Pure red 6.83, pure green 5.33, pure blue 1.37.
- Mid gray 0.01, because the white points agree and gray is safe.
Two people looking at the same hex code on two managed displays see the same color. Two people looking at it on an unmanaged wide gamut panel and an sRGB laptop do not, and the gap is largest exactly where colors are most vivid.
What this means for matching colors
A color matching game is, technically speaking, a metamerism free zone. Both the target and your guess come from the same three primaries on the same panel, so a match on screen is a match in the physical spectrum too. That is unusual, and it is why the scores are meaningful. Nothing in Match or Spot the Difference can fall apart when the light changes, because there is no reflected light involved.
The room around you is a different story. Your adaptation state affects what you perceive even when the emitted spectrum is fixed, which is the mechanism behind color constancy: your visual system is constantly discounting the illuminant it thinks it is under. Playing under a warm desk lamp and playing in daylight will give you slightly different judgements of the same screen color. If you are chasing a leaderboard position, keeping your room light consistent is worth more than most people expect.
The practical rules that fall out of all this:
- Never approve a color match under one light source. Two is the minimum, and daylight plus your actual room bulb is the useful pair.
- Match large samples, not chips. The 5.31 median for the field size effect is real, and it always favours the bigger patch being the honest test.
- If a match has to hold, ask for the same pigment rather than the same color. Identical spectra cannot be metamers. Only identical appearances can.
- For anything on screen, a hex code is only a promise if both ends are color managed. Otherwise you are comparing instructions, not colors.
The counterintuitive one is the third. Everyone tries to solve a mismatch by getting the colors closer. The fix is to get the spectra closer, and those are different goals. Two paints can be a flawless color match and a terrible spectral match, and that combination is precisely what fails when the sun goes down.
The short version
Metamerism is two different spectra landing on the same color, and it exists because three cone types cannot carry 81 numbers worth of spectrum. That compression throws away 78 dimensions, and everything in those 78 dimensions is free to differ without you noticing until the light changes.
Measured, a pair that matches perfectly under daylight drifts a median of 9.13 CIEDE2000 under a tungsten bulb, with 89 percent of pairs visibly failing. Change the field size instead of the light and the median is 5.31. Reproduce a real object on a three-primary display and the match holds for warm colors but breaks by 6 to 7 points in teal and sky blue. Send an sRGB value to an unmanaged P3 panel and the median error is 2.98, rising to 6.83 on pure red.
The short lesson is the one paint shops have been giving out for decades without the arithmetic behind it. Take the sample home. Look at it twice. Colors that agree in one light have made no promises about any other.
If you want to feel the compression rather than read about it, Hex asks you to go from three numbers back to a color, and the Farnsworth Munsell 100 hue test is the classic measure of how finely your own three channels resolve what survives it.