Thin layer method is numerically effective tool to evaluate theoretical ground motions due to large-scale seismic fault, especially when fault parmeters have large complexity or soil property gradually varies in depth direction. In spite of those advantages, there remain several problems in connection with its formulation, e.g., discretization artifacts, artifical reflection of waves impinging at the bottom, and numerical instability when epicentral distance is extremely small. This paper describes some modeling techniques to improve these numerical problems and shows its validation. This method is also applied to ground motion evaluation at near-fault, surface faulting and the 2007 Noto Hanto earthquake.