Species habitat preferences can be obscured when individuals have been recorded in non-core habitats because of dispersal, spillover effects or spatial errors in observation locations. Disentangling the direct effects of the habitats species are observed in from the effects of proximity to other nearby habitats is especially challenging in fragmented landscapes, as many fragmentation metrics are correlated and it is difficult to prove independent effects. In this paper we addressed this issue by comparing a number of models based on predefined ecological theories. We compared models based on quantity of core habitat surrounding observations, proximity to core habitat, or a combination of the two to explain the observed distribution of the saltmarsh inhabiting white-fronted chat (Epthianura albifrons) in coastal New South Wales, Australia. Proximity to core habitat was considered as either Euclidean distance or cost distance, and models were assessed using Akaike's information criterion and the area under the receiver operator characteristic curve on 10 random subsets of the data. We found that all models performed similarly, with the combination of cost distance and the quantity of saltmarsh performing better, but not significantly so. We compared the advantages and disadvantages of different models and also present a model averaged result. Our models suggested that the majority of saltmarshes in New South Wales were too small to have a large effect on probability of occurrence. As climate change is expected to further reduce the amount of available saltmarsh through continued mangrove incursion, coastal populations of the white-fronted chat are expected to come under increasing threat. The conversion of grasslands to urban areas may also increase the effective distance between different populations of the species and reduce gene flow and rescue effects.