-Small (< 50 g) ground-dwelling mammals were sampled as a part of a larger survey of the Pilbara biogeographic region in Western Australia. Environmental attributes infl uencing the probability of occurrence of individual small mammal species were identifi ed using the multivariate adaptive regression splines (MARS) modelling approach. Interpretation of co-occurrence patterns, based on known habitat associations throughout wider ranges in the Pilbara and/or elsewhere in Australia, was used to facilitate comparisons of sub-sets of the species models. Eighteen species were recorded during this study (mean of 3.4 per survey site [range: 0-7]), but only 14 had suffi cient survey records to construct models with reasonable certainty. Complex, non-linear responses were observed for most of the MARS species models, with model performance higher for species with a restricted distribution than the more widely ranging species. Species co-occurrence patterns appeared to be best explained by substrate with three alignments linking species that tend to occur on increasingly sandy, clayey and rocky substrates. Similarly, variables describing the substrate were consistently and strongly represented in the models, indicating that these mammals partition their habitat on substrate type at the local scale in the Pilbara. Climate attributes added little information to the models, although distance to the coast, a surrogate for a range of climatic infl uences, was important for some species. That all species previously known to occur in this region were recorded in the current study suggests that this fauna is still intact, despite a variety of factors (e.g. pastoralism, weed invasion, inappropriate fi re regimes and mining) that have been transforming the region's ecology for more than a century.