During the last decades, the European loess belt has been confronted with a significant increase in environmental problems due to erosion on agricultural land. Spatially distributed runoff and erosion models operating at the catchment scale are therefore needed to evaluate the impact of potential mitigation measures. Expert-based models offer an alternative solution to process-based and empirical models, but their decision rules are only valid for the local conditions for which they have been derived. The STREAM model, which was developed in Normandy (France), has been applied in two Belgian catchments having a similar soil texture, as well as in a catchment of southern France differing by soil, land use and climate characteristics. The performance of hydrological models can be assessed for instance by calculating the Nash–Sutcliffe efficiency criterion (ENS). When applied to Belgium, the model results are satisfactory to good after an adaptation of the decision rules (0.90 < ENS < 0.93 for runoff predictions and 0.85 < ENS < 0.89 for erosion predictions). Given the important environmental differences between Normandy and southern France, the model rules were also adapted for application in the latter environment. Unfortunately, the quality of runoff predictions was insufficient to simulate erosion in southern France. In conclusion, STREAM is a reliable model providing satisfactory runoff and erosion predictions in the regions where hortonian overland flow dominates. Nevertheless, an adaptation of decision rules based on local multi-scale (plot, field, catchment) data is needed, before running the model. STREAM can then serve as a decision support tool to design for instance flood control measures.