Peak stream discharge is a hydrologic parameter that is very important for the determination of flood risk, design of engineering works, and management of water resources. In this study, some 237 stream records across Canada were analyzed using the theory of stochastic processes applied to extreme values. This model, based on partial duration series analysis, was applied to each stream record, considering the time of occurrence of floods to be a Poisson process. In addition, exceedances (values above a given discharge level or truncation level) were considered to be independent random variables identically distributed over a one-year time interval. After this frequency analysis of each stream record, a regionalization of the flood frequency characteristics in Canada was performed using two different approaches: multiple regression analysis and index-flood method. A comparison of the two approaches was carried out by examining mean relative error and root-mean-square error. It was determined that the level of difficulty in applying the stochastic flood model was not the same across Canada. Moreover, error associated with the index-flood method is mainly due to error in estimating low return floods. Key words: flood, partial duration series, regional hydrology, index-flood method, low return flood.