The use of habitat selection models to predict the likely occurrence of wild populations is an important tool in conservation planning and wildlife management. The goal of our study was to build habitat selection models for the dusky dolphin (Lagenorhynchus obscurus) in Golfo Nuevo, Argentina. Random transects were surveyed by boat in 2002–2007. A grid of 1.5 × 1.5-km squares (cells) was constructed for the study area. We characterized each cell by depth, slope, distance from shore, sea-surface temperature, concentration of chlorophyll a, presence–absence of dolphins, and a coefficient of use by dusky dolphins. Models were developed for warm and cold seasons and for all data combined. Data collected during 2002–2005 were pooled to develop the model, and data collected during 2006–2007 were used for cross-validation. Logistic regression with a binomial error structure and a logit-link function were used to relate the presence of dolphins to habitat variables. Models with gamma structure and log-link function were used to relate area use to habitat variables. Models were selected with deviance analysis and Akaike's information criterion. All predictor variables significantly influenced distribution of dolphins, which preferred steep areas at depths of 50–60 m, distance from shore of 3–5 km, and higher values of chlorophyll. In the warm season dolphins preferred colder waters and avoided deeper and warmer areas. Dusky dolphins in Golfo Nuevo are exploited as a tourism resource, and selected models should be considered when deciding the impact of dolphin-watching activities on management.