Colour is an important source of information in the natural world. It can be used for distinguishing and identifying surfaces and objects and separating one region from another. For instance, flowers and grasses in a garden can be distinguished by their colours despite a change in illuminant. Intuitively, the identifiability of surfaces in a scene can be described by their volumes of colour gamuts. But is this approximation of the identifiability accurate? On the other hand, the existence of metamerism in natural scenes shows that colour is sometimes unreliable for surfaces identification. Estimating frequency of metamerism normally requires many comparisons between surface colours to determine their distinguishability under different illuminants. Is there a simpler approach to predict the frequency of metamerism in natural scenes? The aim of this thesis was to address these two questions about the identifiability of surfaces in natural scenes. To answer the first question, the volumes of colour gamuts were estimated over 50 natural scenes under different illuminants. The logarithm of the gamut volume was regressed on the differential entropy of colours. It was found that gamut volume can be an accurate approximation, given a colour difference threshold representing the visual distinguishability within an approximately perceptually uniform colour space. To answer the second question, the frequency of metamerism was estimated over 50 natural scenes with changes in illuminant; and predictive models were constructed based on different combinations of Shannon differential entropies of colours. There was strong dependence of the frequency of metamerism on the combination of the differential entropy and the conditional differential entropy of colours. It means that the frequency of metamerism can be predicted by the informational quantities of the colours in a scene.