© CSIRO 2015. Context Extrapolation of knowledge for threatened taxa between parts of their range that are disconnected and/or ecologically diverse can result in significant sources of error that undermine the effectiveness of conservation efforts. Aims We investigated the risks associated with extrapolation of ecological information across environmental gradients, using the quokka (Setonix brachyurus) as a case study. Information documented in the northern part of its range is currently used to manage this species across its range in south-Western Australia. We examined the suitability of this approach by developing a habitat suitability model (HSM) for the quokka in the southernmost areas of its range and comparing this with existing knowledge for the species. Methods We surveyed 327 sites, representative of a range of ecotypes, for presence/absence of quokkas. Occupancy models were applied to select a subset of habitat variables that best predicted occupancy patterns. Key results Occupancy patterns were influenced by complex vegetation structure, low levels of woody debris and habitat patchiness. HSMs developed for quokkas in the north could not predict occupancy patterns in the south. Significant fragmentation of subpopulations was observed due to patchiness in the availability of suitable habitat. Conclusions The choice of predictor variables in HSMs that are not transferrable between regions could contribute to inappropriate management of habitat for quokkas and an increased risk of local extinctions. In addition, failure to consider processes that affect preferred habitat variables could contribute to the segregation of habitat patches and intervening distances that are too great for successful dispersal, immigration and recolonisation processes. Implications The extrapolation of HSMs between geographical areas can increase the risk of outcomes that are detrimental to the conservation of threatened species. Where such extrapolation is necessary, actions guided by the HSMs should be implemented in a management framework that can detect adverse effects, allow for inclusion of new ecological information and explicitly consider the limitations and assumptions of this approach. In addition, perceptions of habitat fragmentation need to include processes such as fire regimes and feral animals that affect the availability and connectivity of habitat and have the potential to adversely affect population viability.