One of the aims of ensemble prediction systems is to be able to improve forecasts of severe weather conditions. The Met Office has developed a system to generate probabilistic warnings of various severe weather elements from the ECMWF Ensemble Prediction System (EPS), in support of the UK National Severe Weather Warning Service. Since severe weather conditions often occur on small time and length scales, extensive calibration was required to make allowance for the resolution of such events in the ensemble model. It was also necessary to make allowance for uncertainty in timing and location of short-wave events in order to correctly assess the probability of severe weather. Verification results from this system will be presented, along with how these have been used to aid calibration of the system. Initial results based on a relatively small sample indicate some useful skill in the system. Probabilistic skill is found to be greater at a forecast range of four days than at shorter range. Thereafter at longer range the skill declines with lead-time in the normal way. This is believed to be due to the fact that the Singular Vector perturbations used in the EPS are designed specifically for medium-range prediction, and may indicate that the ECMWF EPS is not suitable for severe weather prediction at shorter range.