An ensemble of regional climate models downscaling reanalysis data has been evaluated against observations for the time period 1961-2000. Various aspects of model performance including both their representation of large-scale features and their ability to add value on smaller spatial scales have been considered. A set of metrics has been derived and combined into a performance-based weigthing system that is used in the production of probabilistic climate change projections. Strengths and weaknesses of weighting techniques for RCM ensembles are discussed.