The Amazon River basins present distinct natural and anthropogenic characteristics that influence the structure of stream habitats and their associated biota. The influence of these characteristics can be evaluated through different spatial scales. We aimed to assess the influence (with and without the effect of spatial-geographical factors) of local, macroscale, and land-use variables in the structure of stream fish assemblages of Amazonian catchments with different deforestation levels. A partial redundancy analysis and a reduced metrics model were used to assess these influences. With geographic-spatial effects, we verified that the macroscale and local variables explained the variation in fish composition, and, without the effects, land use also explained the variation in this composition. In the forested catchments, the biota was associated with streams with natural characteristics (e.g., leaf banks). In the deforested catchments, it was associated with land use, sandy catchments with higher soil density (higher capacity of degradation), and less complex streams (fewer leaf banks, more sand). The associated fish have life features linked to these characteristics (e.g., Gymnorhamphichthys rondoni associated with sand). This configuration seems to be a result of both the impact of land use in the catchment (i.e., increased erosion, increased sedimentation) and the naturally sandy constitution of the catchment as well, reflecting the sandy substrate.