Ground‐motion prediction for an earthquake scenario requires evaluation of both the average ground‐motion level and ground‐motion variability due to model uncertainties. The present study attempts to evaluate the ground‐motion variability due to the aleatory variability of the kinematic source parameters. The source models are based on the characterized source model, which consists of several asperities with larger slip and a higher stress drop, as well as the remaining area. We consider the aleatory variability for three source parameters: (1) the location of the asperities, (2) the location of the rupture starting point, and (3) the seismic moment. The first two parameters are chosen randomly based on the fault. The seismic moment is sampled from a normal distribution in which the mean value is given by the relation (with being the fault area) by Irikura and Miyake (2001). The ground motions for a strike‐slip event and a dip‐slip event (in this case, the 2000 Tottori and the 2004 Chuetsu earthquakes, respectively) are simulated by a hybrid approach. The 3D finite‐difference method is used for long periods (), and the stochastic Green’s function method is used for short periods (). Fifty source models were used for each earthquake, and the ground‐motion simulation results obtained (the standard deviations of peak ground velocity, peak ground acceleration, and 5% damped acceleration response) are analyzed at 10‐km‐interval grids. The spatial distribution of the inter‐event variability exhibited distance and azimuthal dependence, which we fitted with a simple regression model using the fault distance and directivity parameters. It was shown that the characteristics of the spatial distribution of varied from short periods to long periods. The spatial distribution of the inter‐event variability for long periods was found to be largely distorted by the complicated subsurface velocity structure.