A new inverse technique is presented here to separate stresses from heterogeneous fault-slip data without a priori information on the stresses nor on the classification of faults according to the stresses. Four parameters are determined by the inversion: one for the shape of stress ellipsoid and three for the direction of stress axes. Accordingly, the inversion is equivalent to the projection of fault-slip data to a point in four-dimensional parameter space. The data are divided into k-element subsets to which inverse technique is applied, where k = 4 or 5 is the optimal choice for the stability of solutions and for the reduction of computation. Significant solutions are identified as the clusters in the parameter space. The technique is demonstrated first by simulated fault-slip data. And, as an example, field data obtained from Miocene fore-arc sediments in western Japan were processed by the method.