Computational models of colour vision assume that the biological function of colour vision is to detect surface reflectance. Some philosophers invoke these models as a basis for ‘externalism’ about perceptual content (content is distal) and ‘objectivism’ about colour (colour is surface reflectance). In an earlier article (Thompson et al. 1992), I criticised the ‘computational objectivist’ position on the basis of comparative colour vision: There are fundamental differences among the colour vision of animals and these differences do not converge on the detection of any single type of environmental property. David R. Hilbert (1992) has recently defended computational objectivism against my ‘comparative argument’; his arguments are based on the externalist approach to perceptual content originally developed by Mohan Matthen (1988) and on the computationally inspired theory of the evolutionary basis for trichromacy developed by Roger N. Shepard (1990). The present article provides a reply to Hilbert with extensive criticism of both Matthen's and Shepard's theories. I argue that the biological function of colour vision is not to detect surface reflectance, but to provide a set of perceptual categories that can apply to objects in a stable way in a variety of conditions. Comparative research indicates that both the perceptual categories and the distal stimuli will differ according to the animal and its visual ecology, therefore externalism and objectivism must be rejected.