A simple iterative learning control scheme is proposed for flexible-joint manipulators to improve tracking accuracy. Insufficient knowledge of robot dynamics and joint flexibility for precise tracking control can be overcome by using this iterative learning law as the manipulation task is repeated. This law guarantees the convergence of the link trajectory to the desired one within bounds, and these bounds can be reduced by properly choosing the learning control parameters. Furthermore, this iterative learning control is computed off-line using link position, velocity and acceleration tracking errors.