Geographic heterogeneity, encompassing both species-environment interactions and interspecific relationships, significantly influences the ecological attributes of wildlife habitat selection and population distribution. However, the impact of geographic heterogeneity on the distribution of target species within predator-prey systems, particularly in human-dominated landscapes, remains unclear. By conducting line transect surveys, utilizing a monitoring network, and applying logistic geographically weighted regression (GWR) in conjunction with generalized linear models (GLM), we examined the spatial heterogeneity of habitat selection by the Amur tiger, Amur leopard, and their main ungulate prey, wild boar and roe deer, in Northeast China. Our results suggest that the factors affecting the spatial distribution of predators are more complex than those for prey. More significantly, the selection coefficients of roe deer and wild boar for certain habitat factors serve as crucial explanatory variables in the Amur tiger and leopard models. Our findings emphasize the importance of spatial non-stationarity in predator-prey habitat selections, and the heterogeneous selection by prey may drive dispersals of large felids across complex road landscapes. This study offers new insights into how to help apex predators cross road barriers by effectively managing prey habitat selection in a landscape dominated by roads, providing valuable guidance for future habitat conservation policies.