The stochastic fluctuating wind pressure field acting on a large-span membrane roof consists of Gaussian and non-Gaussian regions in the sense of statistics. Simulation of this stochastic field is presented based on the zero memory nonlinearity (ZMNL) transformation method. A case study is then given to show that the stochastic wind pressure field samples generated by the proposed method can well represent the specified statistical characteristics of data from the wind tunnel experiment. After that, the wind-induced response analysis is conducted on a membrane roof structure using the Gaussian/non-Gaussian composed wind pressure field samples generated by the proposed method and the Gaussian samples by the traditional method, respectively. Results indicate that response values of some components of the structure induced by the Gaussian samples are lower than those by the Gaussian/non-Gaussian composed wind pressure samples, which means the non-Gaussian characteristics of the stochastic wind pressure should not be ignored. Meanwhile, the total gust response factors of the components caused by the wind excitation are given to meet with the requirement of the designers.