Roads can exert direct and indirect impacts on ecosystems and organisms. In particular, wildlife-vehicle collisions (WVC) may be a considerable threat for populations of certain wildlife species. Despite such threats, there is still incomplete understanding of the factors responsible for high road mortality. Only a few empirical studies have tested the idea that spatial variation of roadkill is affected by environmental characteristics and sociodemographic factors. This study examines the relationships between WVC involving different taxonomic groups (i.e. ungulate, avian, medium mammal, small mammal) and physical and human population characteristics of landscapes by adapting the ecological model, Maxent, to distribution modelling of carcasses resulting from WVC. The WVC-environment relationships are often nonlinear and discontinuous functions, which machine-learning algorithms such as Maxent are well suited to identify. We used 3,252 observations from the California Roadkill Observation System of four taxonomic groups (ungulate, medium mammals, small mammals, and birds) recorded along 1,021 km of 3 state highways in northern California. For each observation, neighborhood characteristics were analyzed acoss13 categories of natural and human-development categories. The Maxent model uses a deterministic machine learning algorithm to optimize environment-species relationships based on maximum entropy. Data are fit using linear and nonlinear functions and different function types can be hinged together. We conducted five runs using a 75-25 split with replacement between training and testing data sets for the four groups. Our results indicated that current carcass-observation locations were explained primarily by total forest area and road density within a 500 m neighborhood. We found that WVC locations were predicted well using environmental variables and human population density together. Moreover, a comparison of projected potential roadkill locations based on environmental factors and human population density among different taxonomic groups revealed substantially different distributions. predicted WVC likelihood varied greatly between the four taxonomic groups. Ungulate WVC was more likely to occur in less developed areas, while avian, medium and small mammal WVC were more likely to occur near existing developed areas. These results indicate potential areas where wildlife populations are at increased risk of coming into contact with traffic and the potential utility of this methodology for modelling current and future distributions of wildlife across landscapes using the Maxent approach.