1. Relating processes occurring at local scale to the natural variability of ecosystems at a larger scale requires the design of predictive models both to orientate stream management and to predict the effects of larger scale disturbances such as climate changes. Our study contributes to this effort by providing detailed models of the hydraulic preferences of 151 invertebrate taxa, mostly identified at the species level. For this purpose, we used an extensive data set comprising 580 invertebrate samples collected, using a Surber net, in nine sites of 2nd and 3rd order streams at one, two or three surveys at each site. We used nested non-linear mixed models to relate taxa local densities to bed shear stresses estimated from FST hemisphere numbers. 2. An average model by taxon, i.e. independent from surveys, globally explained 25% of the density variations of taxa within surveys. A quadratic relationship existed between the average preferences and the niche breadth of taxa indicating that taxa preferring extreme hemisphere numbers had a reduced hydraulic niche breadth. A more complete model, where taxa preferences vary across surveys, globally explained 38% of the variation of the densities of taxa within surveys. Variations in preferences across surveys were weak for taxa preferring extreme hemisphere numbers. 3. There was a significant taxonomic effect on preferences computed from the complete model. By contrast, season, site, average hemisphere number within a survey, and average density of taxa within a survey used as covariates did not consistently explain shifts in the taxa hydraulic preference across survey. 4. The average hydraulic preferences of taxa obtained from the extensive data set were well correlated to those obtained from two additional independent data sets collected in other regions. The consistency of taxa preferences across regions supports the use of regional preference curves for estimating the impact of river management on invertebrate communities. By contrast, the hydraulic niche breadths of taxa computed from the different data sets were not related.