When conventional statistical techniques are used with model‐simulated flows to estimate flood frequencies for different basin scenarios or for regulated flow conditions, the resulting estimates can appear inconsistent with the simulated flood data. In this paper we describe a new approach for flood frequency analysis of model‐simulated flows. Rather than fitting a probability distribution directly to simulated peak discharges, separate distributions are estimated for runoff volumes and for peak discharges conditioned on runoff volumes. The latter estimation is based on rainfall‐runoff modeling using historical rainfall data for the basin of interest as well as data from extreme storms that have occurred in the region. The modeling of extreme events forces consistency in the resulting flood frequency estimates. An example illustrates the proposed method.