Due to the important role of induction motors in industry, reliability in fault detection of these motors is of great importance. In this paper, a method is proposed for increasing the reliability of stator winding fault diagnosis using Dempster-Shafer theory. Out of the measured current and vibration signals, some features are extracted and selected according to the method, and a neural network based on DempsterShafer theory is trained by the selected features. The proposed method is then tested for various fault.Finally it is shown how data fusion can increase the accuracy of fault diagnosis. The system performance is also evaluated on a laboratory motor showing satisfactory results.