Understanding species' associations with physical habitat conditions is a fundamental goal of ecology. For organisms that occupy lotic ecosystems, relationships to streamflow are of particular importance, but these associations are unstudied for most species. We tested the predictability of fish-microhabitat relationships in river shoals (shallow, rocky areas with relatively swift water flow) using a large data set from the Conasauga River in Georgia, USA. Our objective was to assess the consistency of species-specific relationships with flow-dependent variables (depth, velocity, Reynolds number and Froude number) while accounting for other microhabitat variables (e.g. vegetation). We used data from 8285 seine-sets collected during late summer or autumn at 26 sites over 12 years to relate occurrence and counts of 22 fish species to habitat variables using generalised linear multiple regression models.Results showed that microhabitat models explained a substantial amount of the variation in counts for some species, although other species were poorly predicted. We classified 16 species as velocity specialists and nine species as depth specialists, with six species specialised for depth and velocity and three species classified as depth and velocity generalists. The variability in habitat associations that we observed suggests that species will be unevenly affected by anthropogenic activities that alter flows.