Copula function is commonly used in the flood frequency calculation of multidimensional variables because of its flexible construction and wide adaptability. G-H copula is selected to describe the dependence structure of three flood variables, namely flood volume, peak flow and water level. GEV, Logical-normal and PearsonⅢ distribution are considered for fitting the marginal distribution. A two-dimensional G-H copula flood risk assessment model based on peak flow and water level, which are of the highest dependence, is developed to investigate and analyze the characteristics of joint return period (JRP), co-recurrence return period and Kendall return period (KRP). The conditional risk of different combination of flood events is also computed. A three-dimensional G-H copula model is also developed to compare with the bivariate scenario. Qinhuai river basin is chosen as the study area. The result shows that the multi-dimensional joint return period is quite smaller than the co-recurrence return period. Kendall return period is in between of joint and co-recurrence return period but has the minimum error with the single variable return period. The co-recurrence probability of peak flow and water level turns out to be the highest; the smaller the magnitude of peak flow, the less possible for water level to exceed certain threshold. The design value of flood variables in same frequency computed by trivariate G-H copula is larger than the corresponding single variable design value and the difference of other return period with single variable return period is obviously larger than that of bivariate case. The conclusion can be useful for water resource project planning and design and risk assessment of flood disaster in the studied area.