The fault geometry of the 1995 Hyogo-ken Nanbu earthquake in Japan is investigated from the perspective of three-dimensional moment distribution, which in turn is estimated using a linear inversion of strong-motion and geodetic data, without regard to aftershock distribution. The inversion result is evaluated with numerical tests in which intentional modeling errors are given on purpose. In regions shallower than about 9 km, faults on the southwestern and northeastern sides dip toward the southeast and northwest, respectively. Fault segments show a clear stepover at a depth shallower than about 9 km and coalesce into a single fault at a depth of about 14 km. The fault geometry is compared with the distribution of aftershocks that occurred within two hours after the main shock. This shows that aftershocks cluster near the bends and the edges of the main faults. These results imply that the main cause of the aftershocks is the concentration of stresses induced by the main shock.