Abstract: A simple iterative learning algorithm could be used to control the nonlinear dynamical systems by updating the control function from an arbitrary initial trial. This paper improves the convergent condition on the open loop iterative learning schemes posed by Arimoto, M. Togai, Z. Bien, et al., and develops a closed-loop version. It is proved theoretically and demonstrated via simulations that the closed-loop learning scheme is superior to the previous strategies in convergent condition, learning speed and disturbance rejection.