Road-induced habitat fragmentation is one of the greatest threats to large carnivores. Wildlife passes have been used to reduce fragmentation by mitigating the effects of roads as barriers to animal movement. However, direct observations of animals crossing roads are extremely rare and thus indirect methods are necessary to locate crossings. Yet, current methods fail to incorporate the animals' movement behavior and thus have little predictive power. Based on the principles of resource selection functions and state-space modeling, we developed a Bayesian movement model applied to radio-telemetry and GPS data to infer the movement behavior of jaguars Panthera onca as a response to vegetation, roads and human population density in the Mayan Forests of Mexico and Guatemala. We used the results of the model to simulate jaguars moving along a road that bisects the major reserve system in the area. The aim of the simulations was to identify suitable locations for wildlife passes. We found that jaguars move preferentially to undisturbed forests and that females avoid moving close to roads and to areas with even low levels of human occupation. Males also avoid roads, but to a lesser degree, and appear undisturbed by human population density. Simulations reflected these differences: potential crossing sites for females are limited to a strip of a few kilometers, whereas males are able to cross at many different sites. Still, we identified a 1 km strip along the road where the likelihood of crossing for both sexes is highest, ideal for the construction of a wildlife pass. Our study contributes to the ecology of one of the world's least-studied large carnivores and provides a modeling framework that greatly improves the location of wildlife passes. Moreover, our approach can greatly advance region-wide conservation plans for the location of corridors and conservation units.