We present an analysis of the data dependency of surface-wave inversions for source parameters of the 12 July 1993 Hokkaido-Nansei-Oki, Japan, earthquake. This dependency is demonstrated by differences between source mechanisms obtained from different data sets, particularly from Rayleigh waves or Love waves, although inversions of various data sets yield unanimously thrust-fault mechanisms. For the inversion of Rayleigh waves, the mechanism is characterized by a shallow-dipping west-plunging nodal plane, which is consistent with P-wave first-motion data and the distribution of aftershocks close to the hypocenter. However, for the inversion of both Rayleigh and Love waves, the mechanism is characterized by a shallow-dipping east-plunging nodal plane. Several numerical experiments were carried out to test a hypothesis that the data dependency is attributed to the effects of source complexity on the surface-wave radiation, with the source complexity being characterized by tempo-spatial changes in focal mechanism and moment release during the rupture process. The experiments show that source parameters obtained from inversions of surface waves with periods as long as a few hundred seconds can be largely affected by using various models of source complexity proposed previously for the Hokkaido earthquake, and the effects on Love waves tend to be systematically different from those on Rayleigh waves. Our preferred source model is characterized by significant tempo-spatial changes in focal mechanism. The experiments based on this model provide a good explanation for the data dependency, although our analysis cannot completely reject the alternative explanation that attributes the dependency to the manifestation of errors in modeling the surface-wave propagation and/or inversion instability. The model is consistent with the data of body waves and long-period surface waves for a very broad frequency band, implying the existence of weak zones with various fault geometries along the eastern margin of the Japan Sea.