Due to climate change and increasing urbanization, heavy rains-induced floods have occurred more frequently all over the world in recent decades, which is becoming a major deterrent to the sustainable development of social economy. We try to detect risk and return period from both univariate and bivariate cases according to the annual maximum daily streamflow observed at two selected gauges in Yangtze River Basin of China. In this study, copula-based approach is developed for frequency analysis of flood events. The GEV distribution parameter and copula parameter are both estimated through the Maximum Likelihood Estimation method (MLE). Results show that Kendall’s return period can be achieved in the same joint distribution probability level corresponding to the same dangerous area, compared to the traditional return period method (AND/OR method). The definition of Kendall’s return period is more reasonable and can effectively avoid the mistakes of assessing flood events.