The numerical estimation of the geometry of a tsunami source is carried out in two steps; 1) The wave patterns offshore of a bay are computed by the characteristics method from the tide-gage record obtained within a bay, which are used as the fictitious boundary data in the comparison with the numerical results, 2) The best fitting for the reference waves of the numerical results derived from the assumed source deformation consisting of 9 unit sources is determined by the least square method making use of the theorem of superposition. The simplest source model estimated in the case of the 1968 Hyuganada Tsunami is the nearly uniform upheaval of 30-35cm whose area is 80×32km2. The correlation coefficients between the estimated and the observed waves at the stations along the coast of Kyushu and Shikoku are 0.7 to 0.9. This source model is appreciably different from the pattern of the crustal deformation expected from the earthquake mechanism, though the reason is not clear at present.