The study presents the methodology of Bayesian inference and decision making applied to extreme hydrologic events. Inference procedures must consider both the natural or 'modelled' uncertainty of the hydrologic process and the statistical uncertainty due to a lack of information. Two types of statistical uncertainty were considered. The first type is the uncertainty in modelling the hydrologic process, and the second type is the uncertainty in the values of the model parameters. The uncertainty is reduced by considering prior sources of information (regional regression, theoretical flood frequency analysis or subjective assessment) and historical flood data. A case study of determining the optimal size of local flood protection for Woonsocket, considered, using realistic flood damage and cost functions. (Author)