To assist with identifying land for reintroduction, a habitat suitability model (HSM) for black rhinoceros (Diceros bicornis) was developed in the arid Augrabies Falls National Park, South Africa, from records of sightings, feeding trails and dung middens. Logistic regression and Bayesian Information Criterion were employed to construct and select the best HSM from >35 eco-geographical variables. The modelled and the observed distributions of black rhinos did not differ (P = 0.323) and k-fold cross-validation confirmed the model's ability to predict the distribution of independent data. The HSM consisted of five variables: availability and equitability of three preferred foods, distance to roads, habitat heterogeneity, slope and shade. The variables ‘distance to water’ and ‘rockiness’ were also included in the confidence set of models. Only 50% of the study area had a habitat suitability exceeding 11%, but featured 89% of rhino locations. Of 10 vegetation communities, two with high volumes of favourite foods were preferred. Feeding areas with a high density of preferred food plants were also highly preferred. The riverine vegetation community was not preferred, because its abundant browse was not of the preferred species. The apparent avoidance of roads warrants more research and the attention of park managers.