Nearest-neighbour analysis is commonly used to calculate indices of aggregation in groups of animals. It has several problems, however, including lack of data independence and, when studying groups of animals penned at high densities, the difficulty of determining a given individual's nearest neighbour. We describe an entropy-based method to assess the degree of association (or segregation) of groups of animals. We show that this method gives more information and is more: sensitive than the nearest-neighbour technique. An example with a particular experimental situation (mixing groups of lambs) is presented. (C) 2000 The Association for the Study of Animal Behaviour.