Herbicide runoff and the effects of a narrow vegetative filter strip (VFS) were studied on an arable field in the low-lying plains of the Veneto Region (north-east Italy). Cultivated plots were compared with and without a 6 m wide VFS composed of trees, shrubs and grass. Natural and simulated runoff were monitored during 2000 and 2001. Herbicides applied on the field were: metolachlor (2184-2254 g ha(-1)), terbuthylazine (1000-1127 g ha(-1)) and isoproturon (1000 g ha(-1)). The VFS reduced both runoff depth (10.2-91.2%) and herbicide losses (85.7-97.9%) in the monitored rainfall events. Total herbicide loss with runoff was low (0.69-3.98 g ha(-1) without VFS, less than 0.27 g ha(-1) with VFS), but concentrations were sometimes very high, especially of terbuthylazine and isoproturon during the first events after treatment. In these events there was a high probability of exceeding the ecotoxicological endpoint for algae, but the VFS helped to reduce the potential risk. Two VFS effectiveness mechanisms were identified: (i) dilution, and (ii) a "sponge-like" effect, which temporarily trapped chemicals inside the VFS before releasing them. (C) 2005 Elsevier Ltd. All rights reserved.