1. We provide the first theoretical analysis of multihost disease dynamics to incorporate social behaviour and contrasting rates of within- and between-group disease transmission. 2. A stochastic susceptible-infected-recovered (SIR) model of disease transmission involving one to three sympatric species was built to mimic the 1994 Serengeti canine distemper virus outbreak, which infected a variety of carnivores with widely ranging social structures. The model successfully mimicked the erratic and discontinuous spatial pattern of lion deaths observed in the Serengeti lions under a reasonable range of parameter values, but only when one to two other species repeatedly transmitted the virus to the lion population. 3. The outputs from our model suggest several principles that will apply to most directly transmitted multihost pathogens: (i) differences in social structure can significantly influence the size, velocity and spatial pattern of a multihost epidemic; and (ii) social structures that permit higher intraspecific neighbour-to-neighbour transmission are the most likely to transmit disease to other species; whereas (iii) species with low neighbour-to-neighbour intraspecific transmission suffer the greatest costs from interspecific transmission.