AbstractAvailable colour‐discrimination data for surface colours have been analysed to produce chromaticity‐discrimination ellipses for individual colour centres. Applying simulated errors showed that when the distribution of samples around a standard was unsatisfactory, the ellipses obtained were unreliable. A total of 132 reliable ellipses were obtained. The values of 6 and a/b varied systematically over the chro‐maticity diagram, the patterns from acceptability, perceptibility, textile, and non‐textile ellipses being very similar. New experimental data involving over 400 pairs of samples corresponding to 70 of the colour centres and assessed against a grey scale were used to adjust the relative sizes of the ellipses. Overall set factors were used to bring data from different sources onto a common scale. However, even after allowing for different lightness levels, the sizes of the ellipses still appeared to be irregular. Using the new results to adjust the sizes of the individual ellipses from any one group of data, e.g., BFD, DF, or MMB, gave ellipses forming a much more regular pattern. In contrast to the MacAdam ellipses, the smallest ellipses occurred in the blue‐grey region. Our results imply that the relative sizes of the ellipses from the same group of data are in error by a factor of two or more. Comparing colour differences for different colours appears to be much more difficult than has been previously realised. The problem appears to have affected all the major groups of publi