Introduction: Experiments in microcosms were conducted to evaluate how epilithic diatoms communities are affected by the exposition to particular herbicide molecules which are regularly found in rivers at significant concentrations. Specific effects of herbicides on diatoms as revealed by such experiments might be ultimately used as indicators of water contamination. Methods: Before the start of an experiment, glass slides were placed in the upper streams of different rivers in a region of intensive agriculture (Coteaux de Gascogne, SW France). After 3 weeks, the biofilm formed at the surface of the slides and composed mostly of diatoms was removed and equally dispatched into 40 L flasks filled with filtered river water and gradually contaminated with a given herbicide. Glass slides were hung inside each flask and were used to collect weekly biofilm samples. A pump created a surface current between the glass slides to simulate river flow and neon lamps allowed an equal illumination of each microcosm. The biofilm samples were analyzed along the time of experiment for dry weight and chl a content and the composition of diatom communities was determined under microscope (×1000) after sample digestion in H2O2 solution. Results and conclusions: One of our experiments addressed the effect of metolachlor, an herbicide found in spring at high concentrations in the rivers. The tested concentrations were set at 0 (control), 5 and 30 µg/L. The development of diatom biomass at the surface of the glass slides occurred mostly during the second week of experiment to reach 0.3-0.4 µg chla.cm-2 after 14 days (Fig.1). A 20% lower biomass on day 14 suggests a slight reduction of diatom growth at the highest tested concentration. The classification of the different biofilm samples made on day 14 according to the composition of diatom communities does not show any significant variation of the communities between the 3 tested herbicide concentrations (Fig.2). However the occurrence of abnormalities in the morphology of diatom frustules increased with metolachlor concentration, especially for the species Surirella angusta (Fig.3, 8% of clearly deformed frustules at day 14). For the case of metolachlor, the effects of realistic concentrations of the herbicide are more visible at the individual level by microscopic observation rather than at the community level. This experiment supports the use of abnormal frustule frequency as an indicator of toxic contamination.