Nowadays, Human-Machine Systems: HMS has been focused on. In the HMS, a human skill is one of key factors for human machine cooperation. In this paper, a novel human skill evaluation method with the tracking error attractor analysis in a tracking operation task via a joystick is investigated. There are several approaches to evaluate human skills, for example, tracking error measurement or human input characteristics. The proposed skill evaluation method belongs to the human input characteristics. The tracking error attractor consists of proportional and differential tracking error. Then, an estimation index on which proportional or differential tracking error is concentrated by the operator. In the tasks, subliminal calibration, modifying bar dynamics without awareness, and subliminal input filtering, mixing the input command with a human model by neural networks, are applied. The subliminal calibration employed tracking evaluation value, consists of proportional and differential tracking error, to estimate human internal model error. Here, the weight parameters of these two errors are utilize the human skill characteristics. Thus we investigates the attractor by proportional and differential tracking error. Finally, the effectiveness of the subliminal calibration and the subliminal input filtering is discussed by command input characteristics focused on the error attractor.