The dependence of an object's colour on the illuminant chromaticity makes it difficult to use colour as a reliable cue in machine vision applications, particularly in naturally illuminated high dynamic range scenes. To solve this problem the outputs from four logarithmic sensors with different spectral responses can be used to obtain a two dimensional description of an object's chromaticity that is independent of the illuminant. The spectral responses of these four sensors have been optimised. A simple test of colour separability then suggests that using the data from these sensors it is possible to match the ability of the human visual system to separate similar colours. A comparison of the performance of the proposed system when the reflectance and illuminantion data are both changed suggests that readily available data (Munsell reflectance spectra and CIE standard daylight spectra) can be used to design a generic system to separate colours that are described as matching each other. However, for applications that require discrimination between very well matched colours it may be necessary to use an application specific system designed using data relevant to the application.