It is known that the reflectance spectra of both natural and man-made surfaces may be represented efficiently using linear models. A key question, however, is how many basis functions of a linear model are necessary for a given accuracy of representation. The question is ill-posed, however, since it is understood that the number of basis functions required depends to a great extent on the intended application of the linear model. However, in one study it was shown that more than six basis functions were required to ensure that the largest colour difference in the set of spectra was less than 1.0 CIELAB unit and therefore it is reasonable to assume that, for many applications where relatively large patches of spatially uniform colour are present, six of basis functions will be required since CIELAB colour differences of unity or more in such circumstances are known to be noticeable. However, the magnitude of colour difference that would be visible in a complex or natural image is not so well established. A recent psychophysical study demonstrated that although five basis functions produced on average unit error in CIELAB space, original natural images were psychophysically indistinguishable from their linear-model approximations only if there were at least 8 basis functions. The aim of this study is to psychophysically investigate the effect of spatial structure on the number of basis functions required to colorimetrically reproduce spectral images.