The steady-state error associated with the type of self-tuning controller that minimises a cost function incorporating system-input, output and set-point variations is investigated. A method is proposed that eliminates this error without degrading performance.