The northern bettong, Bettongia tropica, is an endangered species of Potoroidae with a restricted distribution in the wet tropics of north Queensland, Australia. The species is only found within a thin strip of sclerophyll forest along the western margin of rainforest. This tight association with rainforest boundaries is predicted to have resulted in population isolation as rainforest contracted during the Pleistocene, though some have proposed that the northern bettong was not present in the wet tropics until the late Pleistocene. The dispersal ability of the species, and of the family, is not known. This study examined gene flow among populations within areas of continuous habitat complemented by a broader analysis of phylogeography. Individuals trapped at each of the four known regions (one region was subsampled at three different sites), were sequenced for 547 base pairs of the mitochondrial DNA (mtDNA) control region and typed for seven microsatellite loci. The mtDNA phylogeny showed congruence with a biogeographical hypothesis, a relatively deep split suggesting historical isolation in separate northern and southern refugia. The two divergent clades were both present within the Lamb Range, indicating an expansion from these refuges and subsequent admixture at one site. mtDNA allele frequencies indicated relatively limited gene flow within the Lamb Range over distances as short as nine km. Tests of population divergence using microsatellites (FST and assignment tests) strongly supported this result. A molecular signal indicative of a recent bottleneck was unexpectedly detected in one of the Lamb Range subpopulations. This lead us to examine the behaviour of the statistics used in this bottleneck test under a linear stepping-stone model with varying migration rates. We found that it may be more difficult to detect molecular signatures for recent bottlenecks under conditions of very low migration rates than for isolated populations and, conversely, that ‘false’ bottleneck signatures may be observed at higher migration rates. The Lamb Range FST estimate clearly fell within the category of potentially ‘false’ bottleneck signals. Despite relatively limited gene flow, evidence for asymmetric dispersal suggests more complicated population dynamics than a simple linear stepping-stone model.