Resumo. Abstract Aim To understand the distribution of fish species and their use of habitat, a range of environmental variables were evaluated as predictors of fish assemblage composition and richness in tropical semiarid aquatic systems. Methods We surveyed the species composition of fish assemblages in semiarid aquatic systems and established their degree of association with the structure of the aquatic habitat. Sites consisted of stream reaches with surface water flow, isolated temporary pools, and man-made reservoirs. Fish sampling was conducted on four occasions during the wet (April and June 2006) and dry seasons (September and December 2006). The correlation between richness and abundance with the habitat structure was evaluated using stepwise multiple regression. Patterns of variation in fish assemblage composition across sites were evaluated using Nonmetric Multidimensional Scaling, and the Indicator Species Analysis was used to determine which species were significant indicators of sites. To establish multivariate correlations and test the hypothesis of local fish composition being associated with the environmental variables we used a Canonical Correspondence Analysis. Results Morphometric variables, stream reach width, stream length and elevation explained 75.6% of the variation in fish richness. Macrophyte cover and overhanging vegetation added to the predictive power of the model equation, where the final model explained 86.9% of the variation in fish richness. Canonical Correspondence Analysis showed a significant relationship between fish composition data and site morphology (altitude, bank slope and littoral depth). Among the water quality, habitat composition and substrate variables, temperature, sand, and gravel showed higher correlation with the CCA axes. Conclusions These results indicated that fish communities assume different structures and compositions across different habitat types following the environmental heterogeneity in dryland aquatic systems.