The fault parameters of the 1994 Shikotan earthquake were determined by inverting all available geodetic data using an iterative linearized least-squares method. A one-fault inversion shows that a shallow-dip fault-plane model does not reproduce the minor deformation observed on Shikotan island, while a steep-dip model can fit all the data well. Multi-fault inversion shows that a fault-plane divided into two subfaults lengthwise is the best fitting model based on Akaike's Information Criteria [AIC> for shallow-dip faulting mechanism. This model yields a very high stress drop, 40 MPa, on the deeper subfault. For a steep-dip model, a fault plane divided into two subfaults sideways gives the minimum AIC. All the computations for models consisting of two to eight subfaults show that the chi-square values of the shallow-dip models are about 10% higher than those of the steep-dip models. Considering the high stress drop, the larger chi-square value, and the requirements of a complex subfault system for the shallow-dip model, we reach the conclusion that the steep-dip model is more suitable to explain the geodetic observations. ¿ American Geophysical Union 1996