For switched reluctance motors, one of the major problems is torque ripple which causes increased undesirable acoustic noise and possibly speed ripple. This paper describes an approach to determine optimum magnetic circuit parameters to minimize low speed torque ripple for such motors. The prediction of torque ripple is based on a set of normalized permeance and force data obtained from numerical field solution for doubly-salient geometries. For that purpose a neural net is trained to extract the data needed to predict the torque produced by a given geometry and excitation at any position of teeth, Hence the static torque curve can be constructed and torque ripple can be found. The accuracy of the approach developed is illustrated by comparing measured and predicted torque for a switched reluctance motor. The optimum parameters for minimum torque ripple conditions are sought using the augmented Lagrangian method. The paper presents the optimization results, and then proceeds to determine the range of a geometric parameters which keep the torque ripple within +/- 10 of the optimum value.