A description of the social network of a population aids us in understanding dispersal, the spread of disease, and genetic structure in that population. Many animal populations can be classified as fission–fusion societies, whereby groups form and separate over time. Examples discussed in the literature include ungulates, primates and cetaceans (Lott and Minta, 1983, Whitehead et al., 1991, Henzi et al., 1997, Christal et al., 1998 and Chilvers and Corkeron, 2002). In this study, we use a heuristic simulation model to illustrate potential problems in applying traditional techniques of association analysis to fission–fusion societies and propose a new index of association: the fission decision index (FDI). We compare the conclusions resulting from traditional methods with those of the FDI using data from African buffalo, Syncerus caffer, in the Kruger National Park. The traditional approach suggested that the buffalo population was spatially and temporally structured into four different ‘herds’ with adult males only peripherally associated with mixed herds. Our FDI method indicated that association decisions of adult males appeared random, but those of other sex and age categories were nonrandom, particularly when we included the fission events associated with adult males. Furthermore, the amount of time that individuals spent together was only weakly correlated with their propensity to remain together during fission events. We conclude with a discussion of the applicability of the FDI to other studies.