In a world with poor biological inventorying and rapid land-use change, predicting the spatial distribution of species is fundamental for the effective management and conservation of threatened taxa. However, on a regional scale, predicting the distribution of rare terrestrial mammals is often unreliable and/or impractical, especially in tropical forests. We apply a recently developed analytic process that integrates density estimation (kernel smoothing), niche-analysis and geostatistics (regression-kriging) to model the occupancy and density distribution of a threatened population of white-lipped peccaries Tayassu pecari in a Brazilian Atlantic forest. Locations (n=45) within a protected area of the Serra-do-Mar state park were obtained from diurnal line transect census (233 km), camera-trapping (751 camera-trap days) and surveys (>626 km) conducted by park rangers. Niche modelling (environmental niche-factor analysis and MAXENT) revealed a restricted niche compared with the available habitat as defined by seven environmental variables. From the occupancy model obtained from regression-kriging, we found that 72% of a 170 km2 protected area is likely to be used by peccaries. We demonstrate that the distribution of large mammals can be restricted within continuous areas of Atlantic forest and therefore population estimates based on the size of protected areas can be overestimated. Our findings suggest that the generation of realized density distributions should become the norm rather than the exception to enable conservation managers and researchers to extrapolate abundance and density estimates across continuous habitats and protected areas.