Understanding landscape structure and the role of habitat linkages is important to managing wildlife populations in fragmented landscapes. We present a data-based method for identifying local- and regional-scale habitat linkages for American black bears (Ursus americanus) on the Albemarle-Pamlico Peninsula of North Carolina, USA. We used weights-of-evidence, a discrete multivariate technique for combining spatial data, to make predictions about bear habitat use from 1,771 telemetry locations on 2 study areas (n = 35 bears). The model included 3 variables measured at a 0.2-km2 scale: forest cohesion, forest diversity, and forest–agriculture edge density, adequately describing important habitat characteristics for bears on our study area. We used 2 categories of unique habitat conditions to delineate favorable bear habitat, which correctly classified 79.5% of the bear locations in a 10-fold model validation. Forest cohesion and forest–agriculture edge density were the most powerful predictors of black bear habitat use. We used predicted probabilities of bear occurrence from the model to delineate habitat linkages among local and regional areas where bear densities were relatively high. Our models clearly identified 2 of the 3 sites previously recommended for wildlife underpasses on a new, 4-lane highway in the study area. Our approach yielded insights into how landscape metrics can be integrated to identify linkages suitable as habitat and dispersal routes.