Mapped environmental classifications are defined using various procedures but there has been little evaluation of the differences in their ability to discriminate variation in independent ecological characteristics. We tested the performance of environmental classifications of the streams and rivers of France that had been defined from the same environmental data using geographic regionalization and numerical classification of individual river valley segments. Test data comprised invertebrate assemblages, water chemistry and hydrological indices obtained from sites throughout France. Classification performance was measured by analysis of similarity (ANOSIM). Geometric regions defined by a regular grid and without regard to environmental variables and a posteriori classifications based on clustering the test datasets, defined lower and upper bounds of performance for a given number of classes. Differences in classification performances were generally small. The ANOSIM statistics for the a posteriori classifications were around twice that of all environmental classifications, including geometrically defined regions. The hydro-ecoregions performed slightly better for the invertebrate data and the network classification performed slightly better for the chemistry and hydrological data. Our results indicate that environmental classifications that are defined using different procedures can be comparable in terms of their ability to discriminate variation of ecological characteristics and that alleged differences in performance arising from different classification procedures can be small relative to unexplained variation. We conclude that definition procedures may have little effect on the performance of large-scale environmental classifications and decisions over which procedures to use should be based primarily on pragmatic considerations.