A fundamental problem in ecological research is to explain large-scale gradients in species richness(1,2). Although many causative agents for this phenomenon have been suggested, the species richness-energy hypothesis has received the strongest empirical support(3-6) : this hypothesis states that higher energy availability provides a broader resource base, permitting more species to coexist. Here we show that the species richness-energy hypothesis applies to North American mammals only over a limited geographical area in which climatic energy levels are low (Alaska and most of Canada), rather than on a continental scale as had previously been accepted(6). In relatively high-energy regions of North America, corresponding to most of the continental United States and southern Canada, we find that mammal species richness is best predicted by topographic heterogeneity and local variation in energy availability, Our results contradict previous studies of large-scale richness patterns that dismissed the importance of habitat heterogeneity(2,7-9), and have implications for climate change research.