Studies have demonstrated negative effects of non-native, predatory fishes on native amphibians, yet it is still unclear why some amphibian populations persist, while others are extirpated, following fish invasion. We examined this question by developing habitat-based occupancy models for the long-toed salamander (Ambystoma macrodactylum) and non-native fish using survey data from 1,749 water bodies across 470 catchments in the Northern Rocky Mountains, USA. We first modeled the habitat associations of salamanders at 468 fishless water bodies in 154 catchments where non-native fish were historically, and are currently, absent from the entire catchment. We then applied this habitat model to the complete data set to predict the probability of salamander occupancy in each water body, removing any effect of fish presence. Finally, we compared field-observed occurrences of salamanders and fish to modeled probability of salamander occupancy. Suitability models indicated that fish and salamanders had similar habitat preferences, possibly resulting in extirpations of salamander populations from entire catchments where suitable habitats were limiting. Salamanders coexisted with non-native fish in some catchments by using marginal quality, isolated (no inlet or outlet) habitats that remained fishless. They rarely coexisted with fish within individual water bodies and only where habitat quality was highest. Connectivity of water bodies via streams resulted in increased probability of fish invasion and consequently reduced probability of salamander occupancy. These results could be used to identify and prioritize catchments and water bodies where control measures would be most effective at restoring amphibian populations. Our approach could be useful as a framework for improved investigations into questions of persistence and extirpation of native species when non-native species have already become established.