Coseismic slip distribution on the fault plane of the 1946 Nankai earthquake (Mw 8.3) was estimated from inversion of tsunami waveforms. The following three improvements from the previous study (Satake, 1993) were made. (1) Larger number of smaller subfaults is used; (2) the subfaults fit better to the slab geometry; and (3) more detailed bathymetry data are used. The inversion result shows that the agreement between observed and synthetic waveforms is greatly improved from the previous study. In the western half of the source region off Shikoku, a large slip of about 6 m occurred near the down-dip end of the locked zone. The slip on the up-dip or shallow part was very small, indicating a weak seismic coupling in that region. In the eastern half of the source region off Kii peninsula, a large slip of about 3 m extended over the entire locked zone. Large slips on the splay faults in the upper plate estimated from geodetic data (Sagiya and Thatcher, 1999) were not required to explain the tsunami waveforms, suggesting that the large slips were aseismic. Two slip distributions on the down-dip end of the plate interface, one from geodetic data and the other from tsunami waveforms, agree well except for slip beneath Cape Muroto in Shikoku. This suggests that aseismic slip also occurred on the plate interface beneath Cape Muroto.