The rupture process of the 1983 Japan Sea earthquake (MJMA = 7.7) is determined by applying waveform inversion of displacement-type, strong motion records based on the Bayesian method. The synthesis of the seismic motions from the main shock is made using the records from the two aftershocks (MJMA = 7.1 and MJMA = 6.1) as empirical Green's functions. The main shock fault consists of two subfaults. The aftershock (M = 6.1) occurring south of the main shock fault is used for the southern subfault empirical Green's function and the aftershock (M = 7.1) occurring north of the main shock fault is used for the northern subfault. In this synthesis, it is not necessary to calculate propagation path and local site effects. Therefore, the information about the source process can be extracted from the comparison between observed seismic motions and the synthetic ones without detailed information on the velocity structure along the whole path from the source to each station. The inversion for the rupture process is attempted for both a line source model and an area source model. In both models, the main shock fault surface is divided into several elements whose sizes are determined from the scaling relations between the main shock and its aftershocks. The slip displacement and the rupture starting time on each element are estimated as model parameters in the inversion. A good convergence is obtained after about 10 iteration steps for the line source case and after 5 iteration steps for the area source case. The results are summarized: (1) the average rupture velocity is approximately 2.5 km/sec for the southern subfault and 2.0 km/sec for the northern subfault, and (2) there is a large slip displacement near the rupture initiation point and the north edge of the fault. Thus, there is considerable heterogeneity in rupture propagation and slip distribution over the fault plane.