Abstract: This paper discusses one kind of neural-network-based controller for path control of a multifingered robot hand. We use a neural network controller during the tracking period and use a normal PID controller during the steady period. Experimental results show that the performance of this kind of controller is good.