Abstract
The idea of a colour space where distance corresponds to discriminability has been fundamental to colour vision research since the 19th century. Despite their long-standing success there is a contradiction between the geometric framework that is typically used in these spaces (a particular application of Riemannian geometry) and a view of the transduction of sensory information as the result of a stochastic process. When this is made explicit, a subtly different approach is suggested which turns out to provide a general, and more complete framework for colour space. It is argued further that not only is a contradiction avoided, but that this framework is both intuitive and of real practical value, in particular for researchers interested in the visual behaviour and ecology of animals.