Abstract: Wind tunnel tests of wind loads acting on the roof of Shenzhen new railway station were conducted in a boundary layer wind tunnel at Hunan University, for the conditions without train and with trains inside the station. In addition, the effects of wind speed amplify in the station were also investigated. This paper presents the results of the mean and RMS wind pressure coefficients, and comparisons of the maximum mean negative wind pressure coefficients and fluctuating wind pressure coefficients for the case of without train and with trains were made. Furthermore, the characteristics of probability density functions of wind pressures measured from typical pressure taps were analyzed, and the peak factors of that were estimated under a probability significance level of 99.999%. On the other hand, the results obtained from numerical simulations were also presented in the paper, and the results were compared with the wind tunnel test data. The outcomes of the combined experimental and numerical study indicate that: (1) The maximum mean negative wind pressure coefficients on the roof occur at the windward leading edge region, where the maximum fluctuating wind pressure coefficients occur also in this region; (2) There are some differences of the maximum mean negative wind pressure coefficients and RMS wind pressure coefficients under conditions with different number of trains inside the station, but such effects on the overall pressure distributions on the whole roof are negligible; (3) There are clearly negative skewed distributions for some pressure taps at the windward leading roof edge and much longer negative tails are observed, which follow NonGaussian distributions; (4) Wind flow was found to be congregated inside the holes at the structure, and the maximum wind speed amplify coefficient is up to 1.34; And (5) the results obtained from the numerical simulations are in good agreement with these of the wind tunnel test, illustrating that the Large Eddy Simulation (LES) approach and Discretizing and Synthesizing Random Flow Generation (DSRFG) technique are effective tools for numerical simulations of wind effects on the complex longspan roof structures.