a b s t r a c tAppropriate restoration and conservation measures require a good understanding of the factors limiting the distribution of species, the presence of steep changes in the distribution along environmental gradients and the effect of environmental interactions on species distribution. We used 12 environmental variables describing connectivity, hydrology, climate and stream morphology, to model the distributions of 17 fish species from 2005 Swedish stream sites that were sampled between 2000 and 2011. Modeling was performed using boosted regression trees and random forest, two machine learning techniques to assess the relationship between species distributions and their environment. Temperature, width and connectivity (minimum distance to lake or the sea and water discharge), were the most important variables explaining changes in species distribution at large spatial scales. Response curves of fitted occurrence probabilities along predictors often showed abrupt changes, however, clear threshold effects were difficult to detect. Our results show also differences across species and even in the outcomes of the two algorithms, implying that a simultaneous assessment of multiple species may provide a better signal of ecosystem change than the use of surrogate species.