This study is part of the AMMA - African Multidisciplinary Monsoon Analysis- project and aims at a better understanding and modelling of the Donga catchment (580 km², Benin) behaviour in order to determine its spatially distributed water balance. For this purpose, we applied the REW concept proposed by Reggiani et al.&nbsp;(1998, 1999), which allows the description of the main local processes at the sub-watershed scale. Such distributed hydrological models, which represent hydrological processes at various scales, should be evaluated not only on the discharge at the outlet but also on each of the represented processes and in several points of the catchment. This multi-criteria approach is required in order to assess the global behaviour of hydrological models. We applied such multi-criteria strategy to the Donga catchment (586 km²), in Benin. The work was supported by an observation set up, undertaken since 1998 consisting in a network of 20 rain gauges, an automatic meteorological station, 6 discharge stations and 18 wells. <P style="line-height: 20px;"> The main goal of this study was to assess the model's ability to reproduce the discharge at the outlet, the water table dynamics in several points of the catchment and the vadose zone dynamics at the sub-catchment scale. We tested two spatial discretisations of increasing resolution. To test the internal structure of the model, we looked at its ability to represent also the discharge at intermediate stations. After adjustment of soil parameters, the model is shown to accurately represent discharge down to a drainage area of 100 km², whereas poorer simulation is achieved on smaller catchments. We introduced the spatial variability of rainfall by distributing the daily rainfall over the REW and obtained a very low sensitivity of the model response to this variability. Simulation of groundwater levels was poor and our results, in conjunction with new data available at the local scale, suggest that the representation of the processes in the unsaturated zone should first be improved, in order to better simulate soil water dynamics and represent perched water tables which were not included in this first modelling study.