Some conceptual models seeking to explain the coexistence of multiple species in hyperdiverse settings predict that species will not be randomly distributed with respect to each other. In stark contrast the 'stochastic geometry' model assumes that a species fine-scale spatial distribution is independent of that of other species in the community. Empirical tests in temperate and tropical forests have provided support for both perspectives. Using point pattern analyses we assessed the prevalence of heterospecific associations between > 10 500 pairs of species and > 3400 pairs of plant functional types (PFTs) in four biodiverse shrubland communities in southwestern Australia. After controlling for first-order effects, spatial associations between species and PFTs were rare, but were most prevalent at the least species-rich of the four sites considered. Individuals tended to have fewer species in their local neighbourhoods than expected under a null model of random relabelling, with this departure most pronounced at the site with fewest species. The consistency of neighbourhood composition experienced by individuals of the same species is, as a result, less than the average under random mixing. Our results demonstrate that the frequency of heterospecific spatial associations is both rare in speciose systems and declines with species richness, and provide further empirical support for the stochastic geometry assumption in species-rich communities. © 2013 The Authors.