This paper describes a formal statistical model underlying the region-of-influence method routinely used in regional frequency analysis of hydrological extremes, and is an improvement to the existing Flood Estimation Handbook (FEH) method for pooled frequency analysis of annual maximum flows in the UK. Specification of a pooling-group method requires three issues to be resolved: how to define hydrological similarity, the size of pooling groups and calculation of the pooled L-moment ratios. Because these issues are interrelated, an exploratory and iterative study has been undertaken before arriving at the final version of the method. Improvements provided by the model are: (1) that it allows an increased weight to be given to a gauged catchment when it is itself the target location and (2) it does not require identification of a homogeneous region, since the expected differences between the L-moment ratios within a pooling group are explicitly accounted for. Using annual maximum series from 602 gauged rural catchments, a comparison of candidate methods shows that the new method performs better than these others, including the FEH method. While the numerical comparison suggests that the improvement is 4%, and thus only minor, arguments are given for why this is a misleading conclusion.