A method of estimating the rotor position of a switched reluctance machine without the need for a rotor-mounted position sensor has been developed. This method takes advantage of the information derived from known phase voltage and current waveforms. The information is fed as the inputs to a neural network, which after being trained, can correctly map the rotor position to its output. The most accurate mapping results were obtained using a Cerebellar Model Articulation Controller (CMAC) neural network. The performance of the neural network has been tested with measured waveforms from a three phase 120 HP switched reluctance motor. It successfully maps the rotor position with an average root mean square error of one tenth of a mechanical degree.