The common cuttlefish ( Sepia officinalis ) communicates and camouflages itself by changing its skin colour and texture. Hanlon and Messenger (1988 Phil. Trans. R. Soc. Lond . B 320, 437–487) classified these visual displays, recognizing 13 distinct body patterns. Although this conclusion is based on extensive observations, a quantitative method for analysing complex patterning has obvious advantages. We formally define a body pattern in terms of the probabilities that various skin features are expressed, and use Bayesian statistical methods to estimate the number of distinct body patterns and their visual characteristics. For the dataset of cuttlefish coloration patterns recorded in our laboratory, this statistical method identifies 12–14 different patterns, a number consistent with the 13 found by Hanlon and Messenger. If used for signalling these would give a channel capacity of 3.4 bits per pattern. Bayesian generative models might be useful for objectively describing the structure in other complex biological signalling systems.