Abstract: The wind pressure acting on roof structures has significant non-Gaussian features. To solve this problem, a new simulation algorithm is proposed for generating sample functions of a multivariate stationary non-Gaussian wind pressure field. A cubic polynomial is used to realize the transformation of a Gaussian process and a non-Gaussian wind pressure process. Then a set of nonlinear equations is derived to determine the parameters of the polynomial. Based on the properties of joint moments of Gaussian stochastic vector, the relation between correlation functions is obtained. Therefore the non-Gaussian wind pressure field is expressed as a function of a fictitious Gaussian vector process, and the Weighted Amplitude Wave Superposition method can be employed to generate the underlying Gaussian processes. Further, the proposed method is illustrated through a wind pressure field simulation of a large-span stadium roof. It is shown that the proposed method is accurate and efficient. Hence the method is advantageous for wind pressure simulation of a large-span roof structure.