Being fully aware of the potentiality of watershed sediment yield and of its capability of being transferred into the hydrographic network is important under different environmental, economic and social aspects. Changes in land use (e.g. conversion from forest or pasture to winter-sown cereals) can increase runoff, exacerbate floods and increase sediment loads resulting from soil erosion into water courses and lakes. Furthermore, an excessive sediment load can determine the deterioration of water quality of rivers and reservoirs and can cause severe sedimentation problems in irrigation channels. In addition to it, sediment can damage fish stocks, while shallow aquatic ecosystems suffer from accelerated infilling and vegetation development (WWF, 2001). Sediment can also damage port areas and tourism places, due to the increased turbidity of water. On the other hand, sediments can play an important role in the conservation of the biological equilibrium and in preventing the shoreline withdrawal. Sediment can also be a valuable economic resource as a supply of minerals and construction materials and as a fertile alluvial deposit for agriculture in coastal areas. At the present time, sediment management is addressed fragmentarily by EU policies and directives and needs a more specific approach. The European Sediment Research Project (SEDNET) pointed out that, unless governments do not take the sediment issue into consideration, there will be a real risk of failing to implement soil and water policy. In Europe there is an urgent need of guidelines at watershed scale in order to achieve a sediment management which is environmentally and economically sustainable. This lack of strategies together with laws not specifically focused on the sediment issue can result in high cost for both the society and the environment. In view of integrating sediment management in the context of the Italian and European policies, the FLORENCE model was developed to predict potential sediment yield from river basins. The quantity estimated by the model is the volume of humid sediment per basin area unit ( units: m³ km^-2 yr^-1). The model derives from the analysis of 59 Italian basin-reservoir systems distributed across Italy. The model was developed as back-error propagation multi-layer neural model by using 10 basin variables to predict sediments volumes accumulated in reservoirs located at the basin outlets. The FLORENCE model variables were determined in GIS environment. To test the FLORENCE model, we used a cross validation procedure by producing 59 neural trial sub-models, each of which considered 58 watershed-reservoir systems, excluding one at a time. The sedimentation values observed in the excluded reservoirs were compared with the sub-model's forecasts. Observed versus sub-model predictions gave a good determination coefficient R²=0.73. The FLORENCE model was then applied to all Italian watersheds (average 5 km² ) and an atlas of potential sediment yield was produced. The model resulted positively validated also by comparing the predicted values of sediment yield with the values of erosion risk estimated by the USLE model and by the PESERA model. The FLORENCE model can be considered more suitable than other distributed models in predicting sediment yield at the watershed scale in Italy. In fact, sediment trapped by reservoirs derives from all sediment sources in the tributary watersheds (soil erosion, mass movements, bank erosion etc.) and not only from sheet and rill erosion.