We proposed a new strong motion prediction method in which both the statistical characteristics of the source, path and site for the Sendai region extracted from observed data and non-linear behavior of soft sediments are considered. This method consists of two major steps. First the strong motions are predicted at the engineering bedrock level on which the non-linear effects of soil are expected to be negligible. The prediction is carried out by using the statistical Green's function method in which the Green's function is calculated by the statistical envelope and spectral model proposed based on borehole data observed in the Sendai region. Second the strong ground motions at the surface of soft sediments are predicted by using these predicted motions as an input to the sediments where its non-linearity should be considered. The 1-D equivalent linear analysis with frequency dependent damping is used in this step. We applied this method to the Miyagiken-oki earthquake of 1978 to verify the method. As a result, we successfully simulated the observed data, the area with high rate of building damage, and the distribution of the rate of overturned tomb stones.