There is an underlying assumption that run-off and rainfall in a given urban catchment are equivalent and, further, to use design rainfall depth as a proxy for run-off in hydrological analyses and calculations. However, when employing this approach, it is difficult to accurately and fully reflect the variability in rainfall characteristics. To address this issue, a method for the copula-based multivariate frequency analysis of rainfall characteristics was proposed by using historical rainfall data (1961-2012) from Guangzhou city. First, continuous rainfall time series were divided into individual rainfall events using the rainfall intensity method. Then the characteristic variables of rainfall were calculated by sampling using the annual maximum method. Finally, a three-dimensional copula was introduced to build a multivariate joint probability distribution model of rainfall characteristics. The results show that the copula-based multivariate analysis is easy to implement and provides reliable results. This approach can be used to analyse the conditional probabilities of variables for different orders of magnitude. It can fully reflect rainfall characteristics, which serve an important reference for urban flood control and drainage planning.