The validation of general circulation model (GCM) land surface schemes requires observations of energy and water fluxes averaged over an area of a typical GCM grid cell (100 km¿100 km) for a time period long enough to include the typical climatic conditions in that area. To create such a consolidated data set, soil moisture and surface evaporation are estimated with a detailed model of the surface energy and water balance. Observations from the Hydrological Atmospheric Pilot Experiment in the Sahel (HAPEX-Sahel) in Niger, West Africa, have shown that rainfall is a key source of spatial variability in the water and energy balance in semi arid regions. The model is therefore run for two wet seasons, forced by observations from a dense rain gauge network over an area of 10,000 km2. Analysis of the estimated surface fluxes indicates that bare soil evaporation has a strong influence on the spatial variability of evaporation during the wet season. On the other hand, leaf-level transpiration is relatively uniform and unstressed by soil moisture for most of the period. Modest spatial variability in transpiration develops through rainfall-induced contrasts in growth rates. The representation in GCMs of these moisture controls on evaporation can be problematic. In this case, where leaf area index is <1, transpiration is well-represented by areal mean surface properties. However, an interactive growth model may be a useful component of the surface scheme when considering the large interannual rainfall variability typical of the Sahel. It is also found that the effect of spatial variability due to subgrid rainfall can not be ignored when modeling interception and bare soil evaporation. Methods of calculating mean fluxes based on mean moisture variables are therefore proposed. Using such a subgrid scheme, total evaporation errors are reduced by 50%. ¿ 2000 American Geophysical Union