Abstract: The novel underactuated robot hand can achieve versatile grasp modes with the ingenious mechanism. This paper focuses on the problem of autonomous grasp of the underactuated robot hand, and divides the problem into two processes: autonomous decision and grasp control. Firstly, the characteristics and grasp modes of the underactuated robot hand are presented. Secondly, proper maps between the fuzzy inputs, including grasp task and object features, and the outputs of grasp modes are established using a fuzzy neural network. According to the grasp modes, the finger orientations are regulated. The control tact based on force sensor feedback is proposed to impose proper forces on all phalanges to achieve stable grasp. At last grasp experiments validate the correctness of autonomous decision and control.