Metapopulation and metacommunity theory may be extremely useful for understanding community composition and species distributions in fragmented landscapes. However, how well particular spatial ecological models represent natural systems is not known and so the general importance of those models is unclear. In three naturally fragmented landscapes in Tasmania, Australia, I sampled beetles from 67 sites, including continuous forest, small forest patches in a sedgeland matrix, streamside forest, and the matrix. Of forty commonly captured beetle species, at least 37% do not form metapopulations of any kind because they do not disperse far enough, disperse too far, or occur in the matrix. Only 7.5% of the species showed distributions consistent with a mainland-island metapopulation, which was surprising given the mainland-island structure of the landscape. These patterns were also consistent with a classic metapopulation. Deterministic metapopulations were probably common, involving at least 32% of species. Most predicted metacommunity patterns were not observed because only subsets of the fauna followed a particular metacommunity model. Seven, mostly flying, species were influenced only by habitat quality, supporting the Species-sorting or Mass-effects models, and the strong distinction between the beetle fauna of Eucalyptus patches and the matrix suggested an over-riding influence of a Species-sorting process. Seven, mostly flightless, species showed a negative relationship with increasing patch isolation, while another six, mostly flying, species showed the opposite pattern, with higher frequency of occurrence in more isolated patches. Both groups were strongly nested. The flying species were probably excluded from less isolated sites through interactions with poor dispersers, providing a novel mechanism leading to a nested distribution. The Patch-dynamics and Species-sorting processes were equally consistent with the pattern, emphasizing that these models are relatively simplistic when confronted with real community data. Additional complexity needs to be built into the models to accommodate the diverse patterns observed in natural communities.