A method for the systematic analysis of precipitation data for the purpose of estimating the probable maximum precipitation is proposed. Empirical distributions of the spatial and temporal sampling errors enable the analyst to estimate the magnitude of the errors attached to the statistics which are used to obtain the extreme precipitation. A comparison of the results from the traditional method with those based on the method presented herein shows differences in both directions which occasionally exceed a factor of two. The emphasis has been on an examination of the data rather than on a search for optimal statistical methods.