Modelling cetacean distributions is crucial to understanding their ecology and to relating use patterns to environmental changes. In the present study, a combination of statistical methods was applied to model the distribution of bottlenose dolphin Tursiops truncatus with 18 physiographic variables around the Island of Filicudi, southern Italy. Principal components and clustering analyses were used to describe the habitat structure derived from mutually correlated predictor variables. Multivariate regression and canonical correlation analyses were used to find critical habitats and core use areas by combining the contribution of 2 response variables: the encounter rate and an index of use calculated according to the spatial behaviour of the dolphin groups. Three critical habitats were identified as distinct combinations of physiographic variables at a 1 km2 spatial scale. Two of these were associated with opportunistic and natural feeding activities. A third, a highly variable topographic habitat located in shallow waters at greater distance from the coastline, appears to serve as a resting/calving habitat by providing defence from anthropogenic pressures. The analysis also estimated an 80% shift from feeding to resting habitats associated with physiographic changes. Since the bottlenose dolphin encounter rate has decreased in recent years, the identification of core areas is useful in the preparation of local marine protected areas for the Aeolian archipelago. This analytical approach to studying dolphin−habitat relationships is relevant for conservation planning as it shows how environmental variability can modify dolphin distribution on a local scale according to the response variables relevant to the species.