Agricultural land use is a primary driver of environmental impacts on streams. However, the causal processes that shape these impacts operate through multiple pathways and at several spatial scales. This complexity undermines the development of more effective management approaches, and illustrates the need for more in‐depth studies to assess the mechanisms that determine changes in stream biodiversity. Here we present results of the most comprehensive multi‐scale assessment of the biological condition of streams in the Amazon to date, examining functional responses of fish assemblages to land use. We sampled fish assemblages from two large human‐modified regions, and characterized stream conditions by physical habitat attributes and key landscape‐change variables, including density of road crossings (i.e. riverscape fragmentation), deforestation, and agricultural intensification. Fish species were functionally characterized using ecomorphological traits describing feeding, locomotion, and habitat preferences, and these traits were used to derive indices that quantitatively describe the functional structure of the assemblages. Using structural equation modeling, we disentangled multiple drivers operating at different spatial scales, identifying causal pathways that significantly affect stream condition and the structure of the fish assemblages. Deforestation at catchment and riparian network scales altered the channel morphology and the stream bottom structure, changing the functional identity of assemblages. Local deforestation reduced the functional evenness of assemblages (i.e. increased dominance of specific trait combinations) mediated by expansion of aquatic vegetation cover. Riverscape fragmentation reduced functional richness, evenness and divergence, suggesting a trend toward functional homogenization and a reduced range of ecological niches within assemblages following the loss of regional connectivity. These results underscore the often‐unrecognized importance of different land use changes, each of which can have marked effects on stream biodiversity. We draw on the relationships observed herein to suggest priorities for the improved management of stream systems in the multiple‐use landscapes that predominate in human‐modified tropical forests.