Why Colours Look Different Under Different Lights

The two jumpers that matched in the shop

Almost everyone has lived the experiment. Two garments, held together under the shop lighting, agree perfectly. Outside, on the pavement, one has turned distinctly greener than the other. Nothing about either garment changed. What changed was the light falling on them, and with it the whole basis on which the eye made its comparison.

This is not a curiosity at the margins of colour work. It is the central practical problem of the field. A colour is not a property that a surface carries around with it like a serial number. It is the outcome of a three-way encounter between a light source, a material that reflects part of that light, and a visual system that responds to what arrives. Alter any one of the three and the perceived colour can move, sometimes by an amount that no professional would accept in a delivery.

What a surface actually does to light

A coloured surface does not emit colour. It subtracts. Daylight arriving on a piece of dyed wool contains energy across the whole visible band, roughly 380 to 780 nanometres. The dye molecules absorb some wavelengths strongly, others weakly, and the remainder is scattered back towards the eye. The record of that behaviour is the spectral reflectance curve: for every wavelength, the fraction of incident light the surface returns, plotted from violet through to deep red.

That curve is the closest thing a material has to a colour fingerprint. It belongs to the sample alone and does not depend on the lamp overhead. Two pieces of fabric with identical reflectance curves will match under every conceivable illuminant, because at every wavelength they are doing exactly the same thing to the light. Colourists call this a spectral match, and it is the only kind of match that carries an absolute guarantee.

Why identical curves are rare

The difficulty is that spectral matches are hard to achieve across different materials and different colourant systems. A shade produced with a reactive dye on cotton and the same shade produced with a disperse dye on polyester are built from different molecules with different absorption behaviour. The dyer can tune the recipe until the two look the same under the agreed reference light, but the underlying curves will cross rather than coincide. They will agree in overall effect while disagreeing wavelength by wavelength.

Metamerism, defined properly

Metamerism is the phenomenon in which two samples with different spectral reflectance curves produce the same colour sensation under one illuminant and different sensations under another. The pair is called a metameric pair, and the failure that appears when the light changes is called metameric failure or, more loosely, flare.

The reason it happens sits in the anatomy of vision. The retina does not perform a spectral analysis. It samples the incoming light with three broad cone types, conventionally labelled long, medium and short wavelength sensitive. Each cone returns a single number, an integrated response across a wide band. A continuous spectral curve, which might be described by dozens of independent values, is collapsed into three. Any collapse of that severity throws away information, and different inputs can produce identical outputs. Two curves that differ substantially can excite the three cone types in precisely the same proportions, and the observer sees one colour.

Change the illuminant and the arithmetic changes. The lamp weights the reflectance curve before the eye integrates it. Where the two curves differ most, a lamp rich in that region will amplify the difference; a lamp poor in that region will suppress it. The cone responses separate, and the match collapses.

The varieties worth naming

  • Illuminant metamerism. The classic case. Samples agree under one light source and disagree under another. This is the failure that costs money in textiles, coatings and packaging, because it is discovered by the customer rather than in the laboratory.
  • Observer metamerism. Two people with normal colour vision disagree about whether a pair matches. Cone sensitivities vary between individuals, and the yellowing of the lens with age shifts short wavelength response measurably. A match approved by one assessor can be rejected by a colleague in good faith.
  • Geometric metamerism. Samples agree at one viewing angle and diverge at another. Metallic and pearlescent finishes, where flake orientation governs how light is returned, are the obvious offenders, but textured textiles show it too.
  • Field size metamerism. A match assessed on a small swatch can fail on a large panel. Cone distribution across the retina is not uniform, which is why colorimetry defines both a 2 degree and a 10 degree standard observer.

Illuminants: the reference lights

Because the light source is half the problem, the industry standardised it. A standard illuminant is a defined spectral power distribution, published as a table of relative energy against wavelength. It is a specification on paper, not a lamp on a shelf, and that distinction matters more than it first appears.

  • D65. Average daylight with an ultraviolet component, correlated colour temperature near 6500 K. The default reference for textiles, plastics and most consumer goods across Europe.
  • D50. Warmer daylight near 5000 K, the reference of the graphic arts and printing industries, and the basis of standard print viewing booths.
  • Illuminant A. A tungsten filament lamp at about 2856 K, heavily weighted towards the red end. Retained as a reference because domestic incandescent light is where a great many metameric pairs break apart.
  • The F series. Fluorescent sources, of which F2 (cool white) and F11 (a narrow band triphosphor) are the most used. Their spiky, discontinuous spectra make them severe tests of a match.

Serious assessment work therefore happens in a light booth containing several calibrated sources, and a match is signed off only after being checked under at least two, conventionally D65 and Illuminant A. A pair that survives both is unlikely to embarrass anyone. Booth lamps age, their output drifts, and hours of use should be logged and the tubes replaced on schedule rather than when they fail.

Colour rendering, and what CRI does not tell you

The move to solid state lighting made the illuminant question urgent again. A light emitting diode produces its white by combining narrow emissions, usually a blue die with a phosphor coating, and the resulting spectrum has pronounced peaks and troughs. Two lamps can share a correlated colour temperature, and therefore look similarly warm or cool on a bare wall, while treating a coloured surface entirely differently.

The colour rendering index, or CRI, was devised to quantify this. The method illuminates a set of standard test colours with the lamp under assessment and with a reference source of the same colour temperature, then measures how far the resulting colours shift. The general index Ra is the average of the first eight shifts, expressed on a scale where 100 means no measurable difference.

The weaknesses of the index

Ra is an average of eight desaturated pastels. It excludes the saturated samples, and in particular R9, the strong red, which is where phosphor converted LEDs most often perform poorly. A lamp can post a respectable Ra of 85 while rendering reds visibly dull, because the deficiency sits in a sample the average never sees. Averaging also hides distribution: a lamp with eight modest errors and a lamp with seven perfect scores and one severe failure can arrive at the same number.

The index is likewise blind to the direction of a shift. Increasing saturation and decreasing it are penalised equally, although observers usually prefer the former. Newer methods address this, notably the TM-30 framework, which uses 99 evaluation samples and reports fidelity and gamut separately so that accuracy and vividness are no longer conflated. Where a specification allows it, asking for TM-30 figures alongside Ra gives a far better picture, and asking for R9 explicitly is the minimum defence when reds matter.

Working with the problem rather than against it

Metamerism cannot be legislated away, but it can be managed. The practical measures are unglamorous and effective.

  • Specify the illuminants in writing. A tolerance is meaningless without the light it applies to. Name the primary source, the secondary check source, and the standard observer, and put them in the purchase specification.
  • Measure the index, not just the eye. Instruments compute a metamerism index, a number expressing how much a pair moves apart when the illuminant is switched. It converts an argument into a figure that can be toleranced.
  • Match with the intended colourants. A recipe developed on the wrong substrate or with a substitute dye class will carry a hidden metameric risk into production, however good it looks in the laboratory.
  • Watch the optical brighteners. Fluorescent whitening agents absorb ultraviolet and re emit blue. Under a UV rich source they lift the white; under a UV poor source they do nothing. A component containing them and one without it will never track each other reliably.
  • Assess components together. Where a finished product combines a woven shell, a moulded trim and a printed label, the parts should be viewed as assembled, in the lighting the product will actually meet.

The engineering position is straightforward once stated. A colour specification that names only a shade is incomplete. A colour specification that names a shade, an illuminant, an observer, a measurement geometry and a tolerance describes something a supplier can hit and a buyer can verify. The jumpers that parted company on the pavement did so because someone, somewhere in the chain, wrote down only the first of those five.