A large-scale field survey was undertaken in the Neste system (South-West France) to investigate the impact of pesticide input on the structure of benthic diatom communities. A total of 18 sites along 4 watersheds (Save, Touch, Sousson, Gèze) plus two complementary sites, located in the Coteaux de Gascogne study area, were sampled during periods of pesticide use over two consecutive years (2005-2006). In this intensive cereal-growing area agriculture is the main source of contamination of the streams. The typology of the diatoms collected was determined using artificial neural networks and the patterns observed in diatom community composition indicated the species influenced by pesticide input, combined with organic pollution. The Phytopixal approach is a means to assess the spatial contamination potential - it was more successful at discriminating diatom assemblages than chemical measurements (total pesticide, maximum concentrations, number of molecules recorded), suggesting that a global estimate of pesticide pollution risk, although rough, would be more appropriate than sporadic analyses.