This paper proposes a new method for knotting a rope with a dual-arm robot system. A knotting task is divided into a series of “steps” taking account for a timing of getting visual information. The start state of each step will be observed and confirmed visually. And an adequate operation from the start state to the end state (i.e. the start state of the next step) is conducted according to the visual information. The method was implemented on a dual-arm robot. Five different types of knotting tasks were realized with the robot system. These knotting tasks have common steps each other and the steps were successfully reused. The experimental result showed that complicated knotting tasks could be realized as a sequence of these kind of steps.