Regional information has proved to be particularly valuable for estimating the higher-order parameters of flood series distributions. This has suggested the use of a hierarchial approach for regional curve fitting, in which higher-order moment ratios, such as the coefficient of skewness and the coefficient of kurtosis, are estimated over larger areas than lower-order moment ratios, such as the coefficient of variation. This philosophy is applied here for the Pearson type III distribution and is combined with an empirical Bayes approach for the estimation of the scale parameter. This work is specifically aimed at estimating flood frequencies and quantiles at sites with short record lengths. Some simulation experiments are reported for an evaluation of the performance of the method, and an application on a set of 59 Ontarian flood series having long record length plus two example stations with short record length is described for illustrative purposes. ¿ American Geophysical Union 1993