Generalized Regression Analysis and Spatial Prediction (GRASP) was used to map the spatial distribution of swordfish (Xiphias gladius) in the South Atlantic. Generalized additive models (GAMs) were used to relate catch to environmental predictor variables. Catch information from 38,000 Brazilian pelagic longline sets from 1980 to 2000 and size frequency data from 5000 longline sets from 1982 to 2000 were obtained from International Commission for the Conservation of Atlantic Tuna (ICCAT). Results highlight the importance of environmental variables for the fishery and for the spatial distribution of different size classes of swordfish (small, intermediate and large). The distribution of swordfish was closely associated with convergence zones (inter-tropical and sub-tropical), especially in the months of greatest convergence intensity. Spatial distribution patterns differed for the three studied size classes. The smallest size classes were found mainly in coastal zones and in areas with a shallow mixed layer (<20 m). In contrast intermediate-sized swordfish were mostly associated with the inter-tropical convergence and mixed layers of more than 20 m depth and large swordfish were often found in the vicinity of the sub-tropical convergence zone and in areas with mixed layers of less than 25 m and greater than 60 m.