Roads impact medium- and large-sized mammal populations through both collisions and barrier effects. This study examined how landscape characteristics influence roadkill occurrences along the Dom Pedro I highway (SP-065), located in the Cantareira-Mantiqueira Corridor, São Paulo State, Brazil. The SP-065 crosses important remnants of the Brazilian Atlantic Forest, a global hotspot for biodiversity. Roadkill records were obtained from the Environmental Company of the State, and land use data were extracted from the MapBiomas platform. We analyzed seven landscape variables (in percentage): native forest, pasture, agriculture, forestry, urban areas, mosaic of uses, and water bodies, considering multiple spatial scales. Mammal species were grouped functionally by home range size and tolerance to anthropogenic environments. In total, 1,418 roadkills of 24 species were recorded, including nine threatened species. Capybaras (Hydrochoerus hydrochaeris) were the most frequently killed species. Generalized linear models showed that, for Group G1 (small home range, common in human-modified areas), roadkills were positively associated with native forest and pasture, and negatively with mosaic landscapes. For Group G3 (large home range, tolerant to anthropogenic areas), agriculture had a positive effect, especially within a 3,000 m radius. For Group G5 (capybara), roadkills increased with pasture and agriculture, while mosaic of uses had a negative effect. Since pasture and agriculture were frequently linked to higher roadkill rates, environmental impact assessments should consider these land-use types when planning mitigation actions. Ultimately, responsibility for roadkill extends beyond highway managers to rural landowners and local governments, as land-use patterns around roads strongly influence mammal movement and mortality.