In the Northern Rockies of the United States, predators like wolves (Canis lupus) and mountain lions (Puma concolor) have been implicated in fluctuations or declines in populations of game species like elk (Cervus canadensis) and mule deer (Odocoileus hemionus). In particular, local distributions of these predators may affect ungulate behavior, use of space, and dynamics. Our goal was to develop generalizable predictions of habitat selection by wolves and mountain lions across western Montana. We hypothesized both predator species would select habitat that maximized their chances of encountering and killing ungulates and that minimized their chances of encountering humans. We assessed habitat selection by these predators during summer using within-home range (3rd order) resource selection functions (RSFs) in multiple study areas throughout western Montana, and tested how generalizable RSF predictions were by applying them to out-of-sample telemetry data from separate study areas. Selection for vegetation cover-types varied substantially among wolves in different study areas. Nonetheless, our predictions of 3rd order selection by wolves were highly generalizable across different study areas. Wolves consistently selected simple topography where ungulate prey may be more susceptible to their cursorial hunting mode. Topographic features may serve as better proxies of predation risk by wolves than vegetation cover-types. Predictions of mountain lion distribution were less generalizable. Use of rugged terrain by mountain lions varied across ecosystem-types, likely because mountain lions targeted the habitats of different prey species in each study area. Our findings suggest that features that facilitate the hunting mode of a predator (i.e. simple topography for cursorial predators and hiding cover for stalking predators) may be more generalizable predictors of their habitat selection than features associated with local prey densities.