A brief summary is given of the history of methods of expressing flood potentialities, proceeding from simple flood formulas to statistical methods of flood-frequency analysis on a regional basis. Current techniques are described and evaluated. Long-term flood records in the United States show no justification for the adoption of a single type of theoretical distribution of floods. The significance and predictive values of flood-frequency relations are considered. Because of the length of flood records available and the interdependence of flood events within a region, the probable long-term average magnitudes of floods of a given recurrence interval are uncertain. However, if the magnitudes defined by the records available are accepted, the relative effects of drainage-basin characteristics and climatic variables can be determined with a reasonable degree of assurance.