Endangered species conservation planning needs to consider the effects of future climate change. Species distribution models are commonly used to predict future shifts in habitat suitability. We evaluated the effects of climate change on the highly endangered mountain gorilla (Gorilla beringei beringei) using a variety of modeling approaches, and assessing model outputs from the perspective of three spatial habitat management strategies: status quo, expansion and relocation. We show that alternative assumptions about the ecological niche of mountain gorillas can have a very large effect on model predictions. ‘Standard' correlative models using 18 climatic predictor variables suggested that by 2090 there would be no suitable habitat left for the mountain gorilla in its existing parks, whereas a ‘limiting-factor' model, that uses a proxy of primary productivity, suggested that climate suitability would not change much. Species distribution models based on fewer predictor variables, on alternative assumptions about niche conservatism (including or excluding the other subspecies Gorilla beringii graueri), and a model based on gorilla behavior, had intermediate predictions. These alternative models show strong variation, and, in contrast to the standard approach with 18 variables, suggest that mountain gorilla habitat in the parks may remain suitable, that protected areas could be expanded into lower (warmer) areas, and that there might be climactically suitable habitat in other places where new populations could possibly be established. Differences among model predictions point to avenues for model improvement and further research. Similarities among model predictions point to possible areas for climate change adaptation management. For species with narrow distributions, such as the mountain gorilla, modeling the impact of climate change should be based on careful evaluation of their biology, particularly of the factors that currently appear to limit their distribution, and should avoid the naïve application of standard correlative methods with many predictor variables.