Short answer
Because its three filters only approximate the colour-matching functions, and the size of the resulting error depends on the spectrum being measured. A display's light is always a mixture of the same three primaries, so the error can be removed with a 3 × 3 matrix found by reading that display's red, green, blue and white with both the colorimeter and a spectroradiometer. The matrix is only valid for displays with the same primary spectra: NIST's tests found that a matrix built on one type of display does not correct another.
A tristimulus colorimeter weights incoming light with three or four filtered detectors. Where a filter's response is a little too high or too low compared with x̄, ȳ or z̄, the reading is wrong by the amount of light that falls at those wavelengths. A broad, smooth spectrum spreads light across the whole filter and the errors partly average out. A display primary is a narrow peak, and it lands on one particular part of the filter curve. Move that peak by a few nanometres, as happens between a phosphor-converted LED backlight, a quantum-dot film and an OLED emitter, and the same colorimeter is wrong by a different amount. Agreement on one panel says little about the next.
The remedy uses a property that reflective samples do not have. ASTM E1455 builds on the fact that everything a display shows is an additive mixture of three primary lights, so the colorimeter's errors on every colour are fixed combinations of its errors on those three. Measure the primaries with a reference instrument as well, and a 3 × 3 matrix maps the colorimeter's readings onto the reference's. Ohno and Hardis at NIST refined the idea into the four-colour method: the matrix is derived from the chromaticity coordinates of red, green, blue and white only. Leaving luminance out of the derivation means that flicker, drift or noise in brightness between the two instruments' readings cannot contaminate the chromaticity correction.
The follow-up study by Ohno and Brown tested how far a matrix travels. Simulating real colorimeters against displays with varied spectra, they found that a matrix made on one type of display and applied to another was not effective, while within one type the remaining chromaticity error stayed within 0.002 in x and y. An experiment on an LCD brought a colorimeter's errors on fourteen colours within the same 0.002 and one ΔE*ab. This is the reason calibration software asks for the panel technology and ships a library of corrections, and the reason a generic correction can be worse than none on a backlight it was not made for. The study's displays were CRTs, LCDs and early OLEDs; the figures belong to them, the principle to every emissive display.
E1455 is explicit about its limits. It applies only where display and instrument are linear, so that the mixture really is additive; it does not deal with bias from timing or geometry, such as a colorimeter and a spectroradiometer viewing different areas or angles of a panel whose colour changes with angle; and it is not a complete measurement procedure. The corrected colorimeter also inherits every error of the reference instrument, including its stray light on narrow primaries. A colorimeter remains the practical tool for speed and low luminance, where a spectroradiometer needs long exposures. For a new panel type, a disputed result or a white point that must be traceable, measure the spectrum directly.
Why: The colorimeter used the same correction on both, so its filter error differs between the two backlights. Observer metamerism between narrow-band whites can add to it.
Fix: Use a correction made for each panel type, or make one with a spectroradiometer on each display.
Why: The colorimeter's filters or the display's emitters have aged, so the pairing the matrix described has changed.
Fix: Rebuild the matrix against the reference instrument on the current display.
Why: The display or the colorimeter is not linear at low levels, so the additive assumption behind the matrix fails there.
Fix: Build and check the matrix at levels where both are linear, and treat near-black readings as less certain.
Each statement is labelled by kind — established fact, a standard’s requirement, observed market data, a convention, or Colourwise’s own interpretation or analysis — with the strength of the evidence behind it.
StandardStrong evidence
ASTM E1455 describes a way to improve the accuracy of tristimulus-colorimeter measurements of self-luminous displays that relies on display light being an additive mixture of three primaries; it is limited to displays and instruments meeting its linearity criteria and does not address bias from temporal or geometric differences between the instruments.
FactStrong evidence
The four-colour matrix method derives a colorimeter's correction matrix from the chromaticity coordinates of a display's three primaries and its white, measured by the colorimeter and by a reference instrument, so that errors in luminance measurement do not enter the correction.
Source: Four-Color Matrix Method for Correction of Tristimulus Colorimeters
FactModerate evidence
In NIST's evaluation, a correction matrix made with one type of display was not effective when other types were measured; with calibration and measurement confined to one type, residual errors were within 0.002 in x and y, and in an experiment on an LCD a colorimeter's errors on fourteen colours were reduced to within 0.002 in x, y and 1 ΔE*ab.
Method: Simulation with spectral data of real colorimeters and displays, plus one LCD experiment.
Caveat: A 1998 study of CRT, LCD and early OLED displays, partly by simulation. The figures should not be assumed for current wide-gamut panels.
Source: Four-Color Matrix Method for Correction of Tristimulus Colorimeters – Part 2
Colourwise interpretationModerate evidence
A colorimeter's error is larger and less predictable on narrow-band display primaries than on broad spectra, because a narrow peak samples one part of each filter's mismatch instead of averaging over it.
Based on: Follows from the colorimeter principle described in the cited instrument literature, and is consistent with NIST's finding that a correction does not transfer between display types with different spectra.
Reviewed 6 October 2026. Colourwise summarises its sources in its own words and does not reproduce standards text or proprietary colour data. Spotted an error? Tell us.