AbstractTwo Regional Climate Model (RCM) projections of changes in extreme precipitation over Europe are assessed and compared. This provides insight into the importance of RCM formulation in representing changes in climate extremes at high spatial resolution. The models concerned are two recent Hadley Centre RCMs, HadRM2 and HadRM3, and are applied at a horizontal resolution of approximately 50 km over Europe, nested within the Hadley Centre coupled Atmosphere Ocean General Circulation Model (AOGCM), HadCM2. The simulation periods are thirty years with fixed concentrations of greenhouse gases representing the climate of 1961–1990 and twenty years representing transient climate change for 2080–2100. The use of common boundary conditions to drive the two RCMs allows us to determine whether their different formulations significantly alter the downscaled projections.The RCM simulations of precipitation extremes are compared with observations from a dense rain‐gauge network over Great Britain, aggregated to the grid used by the RCMs. Both RCMs simulate realistically extreme precipitation occurring over timescales of one to thirty days and for return periods of two to twenty years. In particular, relative errors in the magnitude of extreme precipitation are generally no larger than those in the mean. The two regional models show different patterns of errors for daily precipitation extremes, with the main difference in the western and upland areas of Great Britain where they are underestimated in HadRM2 and overestimated in HadRM3. Change in extremes over all land areas in the domain show increases in intensity everywhere (except for the Iberian peninsula and Mediterranean coast) with most of these significant at the 5% level. Projected increases are greatest for those extremes which are the rarest and shortest duration (i.e. the most intense), both in relative and thus absolute terms. The large‐scale patterns of these changes are very similar in the two RCMs implying they are generally robust to the RCM formulation changes. Given the demonstrated quality of the models this enhances our confidence in the projected changes and suggests that they are mainly conditioned by the large‐scale response in the driving GCM. Copyright Crown Copyright 2007. Reproduced with the permission of the Controller of HMSO. Published by John Wiley & Sons, Ltd