This study focused on characterizing fish assemblages in the Adour-Garonne basin and identifying the relative influences of landscape-scale features on observed patterns in stream fish assemblages. Two different artificial neural network algorithms were used: a self-organizing map (SOM) and a multilayer perceptron (MLP). A SOM was applied to determine fish assemblage types, and a MLP was used to predict the fish assemblage types defined by the SOM. Thirty four species were collected at 191 sampling sites in a major river-system, the Adour-Garonne basin, and topographical factors, namely altitude, distance from source and surface area of drainage basin were measured. Using GIS, land cover types (agricultural land, forests and urbanized artificial surface) were calculated for each site and expressed as percentage of the surface area of basin. These variables were introduced to the MLP and factorial discriminant analysis for the prediction of assemblage types. As a result, the SOM distinguished three fish assemblage types according to the differences of species composition, and the assemblage types were better predicted with landscape-scale features by MLP than discriminant analysis. The percentages of agricultural land and the surface area of a basin showed the greatest influence on assemblage types 1 and 2, and distance from source was the most important factor to determine assemblage type 3.