Bottom-up approaches based on individual behaviour can help to identify key variables influencing populations at larger scales. Instream habitat models have been developed to predict the consequences, for populations in stream reaches, of fish preferences for particular hydraulic conditions observed at the scale of individuals. Conventional instream habitat models (e.g., PHABSIM) predict habitat values for species or life stages in reaches, and their changes with discharge. Despite their worldwide use, they have been subject to continuing criticism and have been mainly limited to site-specific case studies.We ran conventional instream habitat models in 58 French stream reaches dominated by brown trout. Using non-linear mixed effect models, we demonstrated that the outputs of instream habitat models (habitat values for 3 trout life stages and 5 other species) are predictable from average characteristics of reaches (discharge, depth, width and bed particle size). Our models closely reflect variations in habitat values within-reaches (with discharge) and between-reaches. Within-reach changes are linked to the Reynolds number of reaches, while between-reach changes depend mainly on the Froude number at median daily discharge. These two dimensionless variables combine discharge, mean depth and mean width of reaches. Independent model validations showed robust model predictions that are consistent with studies of habitat values for brown trout made in larger streams from western North America. Our results contribute to identifying the main hydraulic variables governing estimates of fish habitat values. They should facilitate habitat studies in multiple streams, at the basin or larger scales, while reducing their cost. They should enhance the biological validation of habitat model predictions, which remains critical.