In this study we propose a framework using neural oscillators for human-robot physical interaction such as “handshaking”. Neural oscillators are used for synchronization and entrainment between human and robot motions. Passiveness of the handshake can be changed by adjusting strength of the synchronization. Joint torque information is taken as an input signal for the neural oscillators, and the neural oscillator generates the desired trajectory of a robot joint. This paper addresses a model structure of the neural oscillator for human-robot physical interaction. The computer simulations of handshaking show the synchronization and the entrainment. Also the experiments with a joint torque sensing robot arm show that a handshake between a robot and a human being is realized by the proposed method. Lastly the validity of the proposed method is examined by a psychological evaluation with a paired comparison method, and it is found that the proposed method is better than conventional impedance control in terms of “Flexible”, “Natural”, “Kind” and “Affinity”.