The wolverine (Gulo gulo) is an uncommon, wide-ranging carnivore of conservation concern. We evaluated performance of landscape models for wolverines within their historical range at 2 scales in the interior Northwest based on recent observations (n = 421) from Washington, Oregon, Idaho, and Montana. At the subbasin scale, simple overlays of habitat and road-density classes were effective in predicting observations of wolverines. At the watershed scale, we used a Bayesian belief network model to provide spatially explicit estimates of relative habitat capability. The model has 3 inputs: amount of habitat, human population density, and road density. At both scales, the best models revealed strong correspondence between means of predicted counts of wolverines and means of observed counts (P < 0.001). Our results can be used to guide regional conservation planning for this elusive animal.