In connection with the workshop 'Comparison of Forest-Soil-Atmosphere Models,' a number of models were applied to the Selling spruce site in Germany. A common set of driving variables and validation data was available for a period of 14 years. Four models, all using the Richards' equation for water transport in the soil profile, were compared and evaluated with respect to differences in various water balance components. In addition, simulations of annual water budgets were compared for five other models with simplified representation of hydrological processes. Major differences were found in the simulated results of the partitioning of evapotranspiration into interception losses, transpiration and forest floor evaporation with respect to both temporal dynamics and total amounts. These differences were found to be mainly caused by different parameterizations and calibration procedures using either tensiometer measurements or chloride concentrations. Generally it was difficult to identify differences caused by the various model concepts because of the more important differences in parameterization procedures. The large variation in results reflects uncertainties in assigning the water balance terms of the Selling site on the basis of the available measurements rather than major differences in the understanding of hydrological processes.