A closed loop neural network based autotuner develops optimized proportional, integral and/or derivative parameters based on the outputs of other elements in the loop. Adjustments are initiated by making a step change in the setpoint which may be done by a user or automatically. A Smith predictor may also be employed.