Island biogeography theory, created initially to study diversity patterns on islands, is often applied to habitat fragments. A key but largely untested assumption of this application of theory is that landscape matrix species composition is non-overlapping with that of the islands. We tested this assumption in successional old field patches in a closely mowed matrix, and because our patches are appropriately viewed as sets of contiguous habitat units we studied patterns of species richness per unit area. Previous studies at our site did not find that diversity patterns on patch ‘islands’ conformed to predictions of island biogeography theory. Our results indicate that when matrix species are removed from the patch samples, diversity patterns conform better to theory. We suggest that classical island theory remains an appropriate tool to study diversity patterns in fragmented habitats, but that allowances should be made for spill-over colonization of ‘islands’ from the ‘sea’.