Short answer
A spectral power distribution (SPD) is a light's full recipe: how much power it emits at each wavelength across the visible range. Colour temperature and CRI are summaries calculated from it, and two lights with the same summaries can still have very different SPDs — which is why they can make the same surface look different.
An SPD plots power against wavelength, usually from about 380 to 780 nm. For comparing lights the absolute level rarely matters, so it is normally given as a relative distribution scaled to 100 at a chosen wavelength (the CIE uses 560 nm). The colour of any surface under that light follows from three multiplications carried out wavelength by wavelength: the SPD, times the surface's reflectance, times each of the three colour-matching functions of the standard observer. Summing each product gives the X, Y and Z tristimulus values. Every colour metric a lighting datasheet quotes — CCT, Ra, TM-30's Rf and Rg — is computed from this curve.
Correlated colour temperature is found from the light's chromaticity, which is itself the ratio of three integrals. Any spectrum that produces the same three integrals has the same chromaticity, and there are infinitely many such spectra. That is the physical reason the site's colour temperature page says the Kelvin figure is necessary but never sufficient: it records where the light sits, not how it got there. Rendering metrics go further by testing the SPD against sample surfaces, but they still compress the curve into a handful of numbers, so the curve itself remains the only complete description.
Light sources fall into three broad shapes. Thermal sources — the sun, incandescent and halogen lamps — give smooth continuous curves; an incandescent lamp's rises steadily towards the red. Phosphor-converted LEDs give a sharp blue peak from the chip followed by a broad phosphor hump, with a trough in the blue-green between them. Gas-discharge sources give lines: fluorescent tubes add mercury spikes on top of their phosphor bands, and low-pressure sodium is almost a single line in the yellow. The table below, computed from the CIE's own data, shows how differently these shapes distribute their power even when their colour temperatures are close.
How the spectrum is sampled matters for spiky light. Reading the fluorescent tables at every tenth nanometre can step over a mercury line entirely: done that way, FL11 computes at about 1,900 K instead of 4,000 K. Averaging each 10 nm band keeps the lines' energy and recovers the right figure, which is how the table below is built.
Three regions decide most practical problems. Deep red, beyond about 640 nm, carries skin, timber, red textiles and meat; a source weak there makes them look dull or brownish however good its Ra. The blue-green trough near 480–500 nm, typical of blue-pumped LEDs, can flatten turquoise and teal. And the violet end below about 420 nm matters for anything containing optical brighteners, which glow only when there is violet or ultraviolet to excite them. Looking for gaps in those three places tells you more about how a lamp will treat a specific object than any single index.
Only a spectroradiometer records an SPD: it disperses the light, usually with a grating, and measures each wavelength band separately. A colorimeter, including the sensor in a phone or a white-balance meter, measures just three filtered channels and so recovers chromaticity and CCT but not the spectrum. Sampling interval matters for spiky sources. The CIE permits 10 nm steps for most colorimetry of smooth spectra, but a mercury line a nanometre or two wide can be missed or overweighted at that spacing, which is why fluorescent and discharge lamps are tabulated at 1 nm or 5 nm and why cheap meters disagree most about them.
| Illuminant | CCT computed from SPD | Peak (10 nm grid) | 380–490 nm | 500–590 nm | 600–730 nm | of which 640–730 nm |
|---|---|---|---|---|---|---|
| Daylight, CIE D65 | 6500 K | 460 nm | 35% | 31% | 34% | 23% |
| Daylight, CIE D50 (graphic arts viewing) | 5000 K | 670 nm | 26% | 32% | 42% | 30% |
| Incandescent tungsten, CIE A | 2860 K | 730 nm | 9% | 24% | 66% | 51% |
| Cool-white fluorescent, CIE FL2 | 4240 K | 440 nm | 29% | 46% | 25% | 8% |
| Narrow-band tri-phosphor fluorescent, CIE FL11 | 4030 K | 610 nm | 27% | 39% | 34% | 7% |
| Phosphor-converted white LED, CIE LED-B3 | 4110 K | 450 nm | 21% | 43% | 37% | 17% |
| Red-green-blue mixed LED, CIE LED-RGB1 | 2840 K | 640 nm | 8% | 36% | 56% | 26% |
Why: Same chromaticity, different SPDs: one has a smooth curve, the other a peaked one with less deep red or a deeper cyan trough.
Fix: Ask for the SPD or a TM-30 report, not just CCT and Ra; compare the 640–700 nm region directly.
Why: Three-channel sensors and coarse spectral sampling misread spiky line spectra.
Fix: Treat app readings of discharge lamps as indicative only; use a spectroradiometer when it matters.
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.
FactStrong evidence
The CIE publishes the relative spectral power distributions of its standard illuminants as open data tables; D65 represents average daylight with a correlated colour temperature of about 6,500 K.
Source: CIE datasets (colour-matching functions, illuminants); CIE (International Commission on Illumination) publications
Colourwise analysisStrong evidence
Sampled at 10 nm between 380 and 730 nm, CIE illuminant A puts about half its power above 640 nm, compared with under a quarter for D65 and under a fifth for the phosphor-converted LED illuminant LED-B3.
Based on: Calculated by Colourwise from the CIE tables for A, D65 and LED-B3 sampled at 10 nm; see the table on this page.
Caveat: Shares are of radiant power in a truncated range, not of visual effect; the eye is less sensitive in the deep red.
Source: CIE datasets (colour-matching functions, illuminants)
FactStrong evidence
TM-30 and CIE 224:2017 compute colour rendition from the light's full spectral power distribution applied to 99 colour evaluation samples, rather than from its colour temperature.
Source: Tutorial: Background and Guidance for Using the ANSI/IES TM-30 Method for Evaluating Light Source Color Rendition (LEUKOS 18:2, 191–231); CIE 224:2017 CIE 2017 Colour Fidelity Index for Accurate Scientific Use
Colourwise analysisStrong evidence
Point-sampling a fluorescent lamp's spectrum every 10 nm can misreport its colour badly: the CIE's FL11 table read at every tenth nanometre gives a correlated colour temperature near 1,900 K, whereas averaging each 10 nm band gives about 4,000 K, its nominal value.
Based on: Colourwise's calculation from the CIE's 1 nm FL11 table and the 1931 2° observer, using McCamy's CCT approximation, once with point samples and once with 10 nm band means.
Source: CIE datasets (colour-matching functions, illuminants)
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.