Abstract: Aiming at the limitations of conventional deduction method of design flood hydrograph, the bivariate joint distribution of flood peak and volume is established by Copula function. The design values of flood peak and volume are jointly simulated by the established joint distribution. Meanwhile, the typical flood hydrographs are chosen based on the similarity between simulated and observed flood values. A bivariate flood risk analysis model is proposed according to the multivariate joint return period. Geheyan Reservoir, on Qingjiang River, is selected as a case and the overtopping risk and ultimate risk corresponding to different flood limited water levels (FLWLs) are estimated. The results show that the FLWL of Geheyan Reservoir can be raised to 193.6m, which will not increase the flood risk if the FLWL dropped to 192.2m before the forecasted flood coming; however the overtopping risk will multiply increase if the FLWL rise to 194.0m, though the ultimate risk remains unchanged. The proposed bivariate risk analysis model can fully consider the randomness and uncertainty of flood process and provide reference information for reservoir operation.