It is highly important in Japan to choose a good site for wind turbines, because the spatial distribution of wind speed is quite complicated over steep complex terrain. We are developing the unsteady numerical model called the RIAM-COMPACT (Research Institute for Applied Mechanics, Kyushu University, Computational Prediction of Airflow over Complex Terrain). The RIAM-COMPACT is based on the LES (Large-Eddy Simulation). The object domain of the RIAM-COMPACT is from several m to several km. First, to test the accuracy of the RIAM-COMPACT we have performed the experimental and numerical simulation of a non-stratified airflow past a two-dimensional ridge and a three-dimensional hill in a uniform flow. Attention is focused on airflow characteristics in a wake region. For this purpose, the velocity components were measured with a SFP (Split-Film Probe) in the wind tunnel experiment. Through comparison of the experimental and numerical results, they showed a good agreement. The accuracy of both of the wind tunnel experiment by the SFP and also numerical simulation by the RIAM-COMPACT were confirmed. Next, we have applied the RIAM-COMPACT to the airflow over real complex terrain. The numerical results obtained by RIAM-COMPACT demonstrated that the changes induced on the wind field by the topographic effect, such as the local wind acceleration and the flow separation, were successfully simulated. The amount of power generation was evaluated in consideration of the correlation with the observational data. Finally, wind simulation of an actual wind farm was executed using the high resolution elevation data. As a result, an appropriate point and an inappropriate point for locating a wind turbine generator were shown based on the numerical results obtained. This cause was found to be a topographical irregularity in front of the wind turbine generator.