Empirical models of habitat selection are increasingly used to guide and inform habitat-based management plans for wildlife species. However, habitat selection does not necessarily equate to habitat quality particularly if selection is maladaptive, so incorporating measures of fitness into estimations of occurrence is necessary to increase model robustness. Here, we incorporated spatially explicit mortality events with the habitat selection of elephants to predict secure and risky habitats in northern Botswana. Following a two-step approach, we first predict the relative probability of use and the relative probability of mortality based on landscape features using logistic regression models. Combining these two indices, we then identified low mortality and high use (primary habitat) and areas of high mortality and high use (primary risk). We found that mortalities of adult elephants were closely associated with anthropogenic features, with 80% of mortalities occurring within 25 km of people. Conversely, elephant habitat selection was highest at distances of 30-50 km from people. Primary habitat for elephants occurred in the central portion of the study area and within the Okavango Delta; whereas risky areas occurred along the periphery near humans. The protected designation of an area had less influence on the proportion of prime habitat therein than did the locations of the area in relation to human development. Elephant management in southern Africa is moving towards a more self-sustaining, habitat-based approach, and information on selection and mortality could serve as a baseline to help identify demographic sources and sinks to stabilize elephant demography. Keywords: African savannah elephant, Anthropogenic features, Habitat quality, Habitat selection, Mortality, Source and sink.