Predicting the composition of benthic macroinvertebrate fauna in rivers is not a trivial task, both because of the number of species to be modelled and because of the complexity of biotic and abiotic relationships that determine their distribution. However, the composition of the benthic macroinvertebrate fauna usually provides very useful insights into the ecological quality of lotic systems, as these organisms are very sensitive to disturbance. Benthic macroinvertebrates are relatively sedentary and long-lived, with life cycle durations ranging from a few months to 2-3 years, and they show a wide range of adaptations to local environmental conditions. They represent a continuous monitoring system of the water body where they are living, but they are also very easy to collect and to identify, at least at an intermediate taxonomic level. Therefore, benthic macroinvertebrates are widely used as biological indicators (Hellawell 1986) and, in particular, they have been used for many years as a source of information for computing several biotic indices that are now used worldwide to assess biological water quality (e.g., Metcalfe 1989, Resh et al. 1996, Lammert and Allan 1999). In this study, the Italian IBE index (Ghetti 1997), derived from the Extended Biotic Index proposed by Woodiwiss (1981) was used as a reference for selecting ecologically homogeneous taxa. Several different biotic indices have been developed, as they had to be suited to ecoregional characteristics in order to provide correct diagnoses of the riverine ecosystem quality. Most indices, however, share the same rationale that is based on the identification of sensitive taxa and on the recognition of the ecological role of other taxa. The main advantage of this approach with respect to more thorough community structure analyses lies obviously in its simplicity. In fact, even people with a limited taxonomic background can be easily trained to carry out rapid surveys aimed at the computation of biotic indices. A more complex approach to the assessment of the ecological status of streams and rivers is based on the prediction of the whole community structure. In the case of benthic macroinvertebrate fauna, different modelling techniques based on ecological knowledge and monitoring data are now available. In the United Kingdom, the work by Wright et al. (1984) led to the prediction of community types on the basis of environmental data by means of a multivariate analysis procedure. This appraoch was then extended and used in the River Invertebrate Prediction and Classification System (RIVPACS) (Wright et al. 1993b), which provides estimates of the ecological quality at a given site by comparing the observed macroinvertebrate fauna composition with the expected one. The RIVPACS approach has also been adapted to other ecoregions. For instance, the Australian River Assessment Scheme (AUSRIVAS) (Simpson and Norris 2000) is based on the RIVPACS approach, although it has been expanded and adapted to each Australian ecoregion. Another method that is closely related to RIVPACS and AUSRIVAS is the benthic assessment of sediment (BEAST) (Reynoldson et al. 1995), that is based on quantitative data about macroinvertebtare fauna instead of presence/absence data only. Even though the RIVPACS approach proved to be very effective, it has limits related to the non-linearity, complexity and dynamic nature of biotic responses to environmental characteristics. Moreover, the development of an assessment system based on the RIVPACS rationale requires a considerable amount of work and thorough statistical analyses. A new generation of empirical techniques for analysing and modelling complex ecological data in a more simple and straightforward way is now emerging. Among these new modelling methods Artificial Neural Networks (ANNs) play a relevant role and represent a useful tool when relationships among data are unknown and/or non-linear. ANNs learn from examples and do not require a priori theoretical models, nevertheless they are able to model complex temporal and spatial patterns and to reproduce the behaviour of very complex systems (Recknagel and Wilson 2000). During the last 10 years, ANNs have been applied to various ecological fields (see, for instance, Lek and Guegan 2000), including studies relating community characteristics with environmental variables (e.g., Chon et al. 1996, Recknagel 1997, Recknagel et al. 1997, 1998, Guégan et al. 1998) and modelling habitat suitability (e.g., Paruelo and Tomasel 1997, Ozesmi and Ozesmi 1999). As for the particular case of macroinvertebrate fauna, Pudmenzky et al. (1998) and Walley and Fontama (2000) recently developed ANN approaches that are aimed at the same goals and ecoregions as AUSRIVAS and RIVPACS respectively. Our study was focused on a benthic macroinvertebrate data set provided by the Latium Regional Environmental Protection Agency and it is aimed at testing different strategies for modelling the presence or absence of macroinvertebrate benthic taxa on the basis of environmental variables, using ANN models.