Flood damage computed from flood-frequency curves fitted mathematically to observed annual peak flows or estimated by regression with basin characteristics, is a biased estimator of flood risk. The inaccuracy inherent in any flood-frequency curve increases the annual premium that would have to be charged to break even over a large number of projects. For ungaged sites where the population of annual peaks can be assumed to follow a log-Pearson Type III distribution, the true risk is evaluated by relating it to the standard error of estimate of the regression used to define the flood-frequency curve. In view of this relation between bias and error, the accuracy of all procedures used in evaluating flood frequency should be appraised in terms of standard error so that the proper flood risk can be obtained.