Short answer
The visual system rescales each cone type's sensitivity to the prevailing light, so a white page still looks white under a warm lamp — chromatic adaptation. Colour science models it with transforms from von Kries's simple scaling to the CAT16 transform in the CIE's current appearance model. It is incomplete, slow, and breaks down in dim light, where rod vision takes over and the Purkinje shift darkens reds.
The core idea, due to Johannes von Kries, is that each of the three cone types adjusts its own gain so that its response to the illuminant's white is held roughly constant. Under warm incandescent light, long-wavelength cones are strongly stimulated by everything, so their gain falls; short-wavelength cones gain up. Divide each cone signal by its response to white and the illuminant largely cancels out. This 'von Kries' scaling explains most everyday constancy and is still the skeleton of every modern chromatic adaptation transform. It also explains its own limits: it can only rescale three channels, so it cannot undo a spectral problem such as metamerism, where the three signals themselves have changed relative to each other.
Modern transforms apply the scaling in a sharpened cone-like space, chosen so that predictions match experiments on corresponding colours. The Bradford transform is widely used in colour management; CAT02 was embedded in the CIECAM02 appearance model; and CAT16, introduced with CAM16 in 2017 to fix computational problems in CIECAM02, is part of CIECAM16, which the CIE published in 2022 as CIE 248 and which supersedes CIECAM02. These models also include a degree-of-adaptation factor, because adaptation is often partial. The metamerism explorer and the tables in this section use a Bradford transform to show what an adapted eye would see.
Adaptation is not instant — the warm tint of a newly entered room fades over moments rather than vanishing at once — and it is incomplete at low light levels and for strongly coloured illuminants: a room lit by 2,700 K lamps looks neutral but still reads warm when you think about it, and very orange sodium light never looks white. People can usually still tell the colour of a room's light coarsely, partly because they move between spaces and see mixed sources. Mixed lighting defeats adaptation outright: the eye can only adapt to one white at a time, so a window-lit wall beside a lamp-lit one will always look tinted on one side.
A camera has the same problem with mixed light, which is why one white-balance setting cannot fix a scene lit by both daylight and tungsten.
Below roughly twilight levels, the cones hand over to the rods, which are more sensitive but come in one type and cannot distinguish hue. Daytime (photopic) sensitivity peaks near 555 nm; night-time (scotopic) sensitivity, standardised by the CIE as V′(λ), peaks near 507 nm and is almost blind to deep red. So as light fades, reds darken faster than blues — red flowers turn black at dusk while blue ones seem to glow — and a moonlit scene looks bluish although moonlight is not blue. In between lies mesopic vision, where both systems contribute; street lighting and dim galleries often sit here, and the CIE publishes mesopic efficiency functions for it.
| Regime | Receptors | Peak sensitivity | What happens to colour |
|---|---|---|---|
| Photopic (daylight, bright interiors) | Cones | about 555 nm | Full colour vision; adaptation works best |
| Mesopic (dusk, street lighting) | Cones and rods | Between the two | Colour weakens; reds darken relative to blues |
| Scotopic (moonlight, starlight) | Rods | about 507 nm | No hue discrimination; scenes look blue-grey |
Why: Your eye adapted to the warm lamps; the camera's white balance did not.
Fix: Set white balance from a grey card under the room's light, or judge by eye rather than by photo.
Why: Purkinje shift: rod-dominated vision is insensitive to long wavelengths.
Fix: Judge colour only at photopic light levels; do not choose outdoor colours at twilight.
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
CIECAM16, published by the CIE as CIE 248:2022, supersedes CIECAM02 (CIE 159:2004); it is simpler while matching its predictive performance for visual data.
Source: CIE 248:2022 The CIE 2016 Colour Appearance Model for Colour Management Systems: CIECAM16
FactStrong evidence
CAM16 and the CAT16 chromatic adaptation transform were introduced in 2017 to overcome computational failures of CIECAM02.
Source: Comprehensive color solutions: CAM16, CAT16, and CAM16-UCS
FactStrong evidence
The photopic luminous efficiency curve peaks at about 555 nm and the scotopic curve at about 507 nm; scotopic vision is mainly rod vision and is very insensitive to red.
Source: HyperPhysics (optics, atmospheric optics and vision pages); CIE spectral luminous efficiency for scotopic vision, V′(λ) (CIE 018:2019 dataset)
FactModerate evidence
Despite effective chromatic adaptation, room occupants can usually still judge the chromaticity of a light source coarsely, partly because they see mixed sources and move between spaces.
Reviewed 29 September 2026. Colourwise summarises its sources in its own words and does not reproduce standards text or proprietary colour data. Spotted an error? Tell us.