Based on the Non-Gaussian Simulation of multivariate stochastic processes method, one prediction method for wind pressure extreme value is proposed. The wind pressures time histories of several taps on opening roof are simulated by Multivariate Non-Gaussian Simulation method based on wind tunnel test data. It is shown that the power spectra density, coherence, deviation, skewness, kurtosis of simulated non-gaussian time histories are very close to the destination. Then the peak factors of wind pressures on opening roof are predicted from Multivariate Non-Gaussian Simulation time histories for several times by typical extreme value theory, and the results are compared with that of several general methods. It is shown that Davenport method overestimated the positive peak factor by 60% whose skewness is negative, and underestimated the negative peak factor by 43%; Sadek-Simiu method overestimated the peak factor by 50% which have high kurtosis; the proposed method can predict the peak factor effectively, and the overall error is smallest.