Predicting sediment yield at the catchment scale is one of the main challenges in geomorphologic research. The application of both physics-based models and regression models has until now not provided very satisfying results for prediction of sediment yield for medium to large sized catchments (c. >50 km2). The explanation for this lies in a combination of the large data requirements of most models and a lack of knowledge to describe all processes and process interactions at the catchment scale. In particular, point sources of sediment (e.g. gullies, mass movements), connectivity and sediment transport remain difficult to describe in most models. From reservoir sedimentation data of 44 Italian catchments, it appeared that there was a (non-significant) positive relation between catchment area and sediment yield. This is in contrast to what is generally expected from the theory of decreasing sediment delivery rates with increasing catchment area. Furthermore, this positive relation suggests that processes other than upland erosion are responsible for catchment sediment yield. Here we explore the potential of the Factorial Scoring Model (FSM) and the Pacific Southwest Interagency Committee (PSIAC) model to predict sediment yield, and indicate the most important sediment sources. In these models different factors are used to characterize a drainage basin in terms of sensitivity to erosion and connectivity. In both models an index is calculated that is related to sediment yield. The FSM explained between 36 and 61 per cent of the variation in sediment yield, and the PSIAC model between 57 and 62 per cent, depending on the factors used to characterize the catchments. The FSM model performed best based on a factor to describe gullies, lithology, landslides, catchment shape and vegetation. Topography and catchment area did not explain additional variance. In particular, the addition of the landslide factor resulted in a significantly increased model performance. The FSM and PSIAC model both performed better than a spatially distributed model describing water erosion and sediment transport, which was applied to the same catchments but explained only between 20 and 51 per cent of the variation in sediment yield. Model results confirmed the hypothesis that processes other than upland erosion are probably responsible for sediment yield in the Italian catchments. A promising future development of the models is by the use of detailed spatially distributed data to determine the scores, decrease model subjectivity and provide spatially distributed output. Copyright © 2006 John Wiley & Sons, Ltd.