Stochastic weather generators can be used to produce long periods of synthetic weather records from a limited amount of input data. They provide a useful tool for supplementing existing climate data for risk assessment and decision support systems. They can be adapted to climate change studies where climate scenarios are available from General Circulation Models on a coarse grid. The climate scenarios from climate models are frequently not appropriate for direct input into impact models. Downscaling techniques, using stochastic weather generators, have been widely used to bridge the gap between climate and impact models. A number of studies have demonstrated the failure of data produced with stochastic weather generators to mimic the statistical properties of observed daily data or to reproduce weather variability and the magnitude of extreme events. Research is continuing to develop model improvements which address these limitations. This study examines the distribution of extreme values of daily temperature and precipitation for selected Canadian climatological stations with period of record of 90 or more years and compares the corresponding distributions generated with a stochastic weather generator. The objective is to evaluate how well the extreme value distributions in the generated data correspond with the observed and assess how well standard input parameters used in the weather generator can account for trends in variability observed in long term weather series. The effect of different approaches used to specify the probability distribution within the weather generator is examined. It has been shown that generating daily temperatures from an unbounded normal distribution can result in physically improbable values. This study examines the use of common distributions as well as empirical distributions in order to establish how best to improve the correspondence with the observed. The use of theoretical extreme value distributions in a stochastic weather generator to improve the simulation of extreme events is tested on selected Canadian climate station data. The application to climate change studies is addressed.