Many wildlife populations are either naturally, or as a result of human land use, patchily distributed in space. The degree of fragmentation-specifically the remaining patch sizes and habitat configuration-is an important part of population dynamics. Demographic stochasticity is also likely to play an important role in patchy habitats that host small local populations. We develop a simulation model to evaluate the significance of demographic stochasticity and the role fragmentation plays in the determination of population dynamics and the risk of extinction of populations on habitat patches. Our model is formulated as a Markov-chain stochastic process on a finite, spatially explicit array of patches in which probability of successful dispersal is a function of interpatch distance. Unlike past work, we explicitly model local population dynamics and examine how these scale up to the entire population. As a test case, we apply the model to the American pika (Ochotona princeps) population living on the ore dumps in the ghost mining town of Bodie, California. This population has been studied nearly continuously for over four decades and has been of conservation concern as the southern half of the population declined precipitously beginning in 1989. Our model suggests that both the specific configuration of habitat and landscape heterogeneity are necessary and sufficient predictors of the eventual extinction of the southern constellation of patches. This example has important implications, as it suggests that fragmentation alone can lead to regional extinctions within metapopulations.