We use a simple test to evaluate the null hypotheses that desert rodent communities are composed of species assembled at random with respect to body size. This hypothesis is unequivocally rejected for species in the granivore guild: species of similar size (body mass ratios <1.5) coexist less frequently in local communities and overlap less in their geographic distributions than expected on the basis of chance, suggesting that their co—occurrence is precluded by interspecific competition. When granivore species and members of other guilds are combined in the same analysis, the patterns apparent in granivores diminish or disappear. These results indicate that ability to detect community structure depends to a large extent on thorough knowledge of the organisms and choice of appropriate statistical tests. We suggest that recent claims that many communities are indistinguishable from random associations of species are premature.