Abstract: The coefficients and power spectra of wind pressure play important roles in computing wind-induced dynamic responses of long-span roofs. In this paper, a neural network method is used to predict the mean and fluctuating wind pressure coefficients and the power spectra of the fluctuating wind pressures on a real long-span roof based on the limited data from the wind tunnel test. The predictions are in good agreement with the experimental data, indicating that the present method is effective.