Spatial patterns of near-surface wind speeds and resulting loss potentials associated with severe winter storms were investigated in multi-model simulations with regional climate models (RCMs), driven by ERA40 re-analyses. The benefits and limitations of dynamical downscaling for windstorm loss calculations were explored, including a quantification of the performance of the multi-model ensemble as a whole and the systematic investigation of the influence of model selection on the ensemble results. A comparison of the wind fields in the different models revealed both systematic biases in individual RCMs and model-specific anomalies over mountainous regions. Further, a storm loss model was applied to the RCM wind fields, and the calculated losses were validated against observed annual insurance loss data available for Germany. Generally, a distinct advantage from dynamical downscaling was obvious. However, all RCMs failed in realistically simulating 1 specific major event. If this particular event was excluded from the considerations, almost all simulations revealed high correlations (above 0.8) with observed losses, comparable to losses calculated directly from the large-scale reanalysis wind field. For the best performing models, considerably higher loss correlations up to 0.95 were obtained, suggesting that the high-resolution RCMs exceeded the value of assimilation in the driving data for the area considered. Combining calculated losses from the individual RCMs into a multi-model ensemble, the performance of the ensemble mean was as good as the performance of the best single model. Examining all possible sub-ensembles, we found that generally a higher minimum performance was obtained with a larger number of ensemble members, whereas the maximum performance was hardly affected by the ensemble size.