Short answer
Three different numbers describe it. Repeatability is how closely one instrument agrees with itself on the same sample; inter-instrument agreement is how closely two instruments agree; and uncertainty is the stated range within which the true value probably lies, built from every source of error. A good instrument may repeat to a few hundredths of a ΔE unit yet disagree with another model by a whole unit or more — so tolerances tighter than instrument agreement cannot be enforced between parties.
A reflectance spectrophotometer is calibrated at the start of a session by measuring a white tile of known spectral reflectance (and often a zero reference such as a light trap). That sets the top and bottom of the scale for each wavelength. The tile's values are themselves traceable, through the manufacturer's laboratory, to a national metrology institute such as NIST or PTB. The chain matters: a dirty, scratched or wrongly assigned tile shifts every measurement, and a tile left in sunlight or handled can drift. Many instruments also check wavelength accuracy and linearity against reference tiles at service intervals.
Short-term repeatability — measuring the same spot of a stable white tile many times — is what instrument data sheets quote, and it is usually very small. Real samples are less forgiving: placement, pressure, texture, temperature and sample non-uniformity add variation, so repeated measurements of a real print or fabric differ more than of the tile. Reproducibility over days and between operators is larger again. ISO 12647-7 recognises this for proofs: it limits variation across one proof sheet (standard deviation under 0.5 in L*, a*, b*) and day-to-day reproducibility of the proofer separately from the accuracy tolerance.
Two instruments of the same model agree closely; two different models can disagree by a ΔE00 or more on the same sample because of different optics, bandwidths, lamps and UV content. The contract-proofing standard says outright that tolerances below 2.5 ΔE00 are not practical because of poor inter-model agreement, and recommends halving tolerances when one instrument measures both reference and sample. Research on graphic-arts instruments found that the M1 measurement condition greatly improved agreement between models on brightened papers compared with legacy M0 devices, showing that part of the historical disagreement came from undefined UV rather than from the instruments' precision.
Uncertainty is the metrologist's way to combine all of this into one statement. NIST's guidance divides components into Type A (evaluated statistically from repeated measurements) and Type B (evaluated by other means — calibration certificates, instrument specifications, known effects such as temperature), combines them as a combined standard uncertainty, and multiplies by a coverage factor, commonly k = 2, to give an expanded uncertainty at roughly 95 % confidence. A result reported as 'ΔE00 = 1.2 ± 0.4 (k = 2)' can be judged against a tolerance; a bare '1.2' cannot, because a pass by 0.1 inside the uncertainty is not really a pass.
Why: Different instrument models whose disagreement is comparable to the tolerance.
Fix: Exchange a physical reference and compare both instruments on it; set tolerances above inter-instrument agreement.
Why: Instrument or lamp warming up, or the white tile has been contaminated.
Fix: Recalibrate at set intervals, keep the tile clean and covered, and log a check-standard reading.
Why: The result was within measurement uncertainty of the tolerance limit.
Fix: Report uncertainty and treat results within it of the limit as inconclusive.
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
NIST Technical Note 1297 classifies uncertainty components as Type A (evaluated by statistical methods) or Type B (evaluated by other means), combines them into a combined standard uncertainty, and expresses an expanded uncertainty using a coverage factor such as k = 2.
StandardStrong evidence
ISO 12647-7:2016 states that specifying ΔE00 tolerances lower than 2.5 is presently not practical because of poor inter-model agreement, and recommends halving tolerances when the same instrument measures both sets of values.
Source: ISO 12647-7:2016 … — Part 7: Proofing processes working directly from digital data
FactModerate evidence
A comparison of ten commercial spectrophotometers found greatly improved inter-model agreement between ISO 13655 M1-compliant instruments compared with earlier hand-held instruments measuring in legacy M0 mode.
Caveat: A single study of ten instruments on graphic-arts substrates.
StandardStrong evidence
ISO 12647-7:2016 limits variation across nine points on one proof to a standard deviation below 0.5 in each of L*, a* and b*, and limits day-to-day change in control patches to 2.0 ΔE00.
Source: ISO 12647-7:2016 … — Part 7: Proofing processes working directly from digital data
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.