Although the conservation of endangered species often implies the definition of priority areas for conservation, detailed information on their distribution patterns is seldom available over a large geographic range. The present paper explores the performance of an alternative data analysis approach, artificial neural networks, for assessing distribution patterns of endangered mammals when data are scarce and noisy. This approach was applied to identify wolf occupancy in Portugal based on information on wolf depredation and inquiries. Artificial neural networks were able to discriminate successfully between areas sporadically used by vagrant individuals and areas occupied by resident wolves, with a low estimated prediction error (around 1%). Only 33% of the wolf range is regularly occupied, being fragmented in five nuclei. These nuclei are surrounded by more disturbed marginal areas, with a source-sink dynamic between nucleus and marginal areas. Selection of priority areas for wolf conservation in the context of the Natura 2000 Network increased twofold the proportion of wolf regular range within protected areas. The good performance of artificial neural networks in assessing wolf distribution patterns suggests that this approach may be applied to other species where detailed records of distribution are limited.