The pitch angle control of the underwater towed vehicle during its working process has always been an important part of underwater towing system attitude control. In this paper,a direct adaptive controller for underwater towed vehicles based on RBF neural network is designed,using the characteristic of local infinite approximation to nonlinear functions of RBF neural network in closed-loop system. With the cooperation with the traditional PD controller,the output of RBF neural network replaces the nonlinear uncertainties in the dynamics model of underwater towed vehicle. No prior off-line learning phase,the adaptive controller learns to update neural network weights on-line. Control law and neural network weight update law are proved stable by using Lyapunov theory,and the tracking error converges to 0.This paper compares the performance of the controller with the traditional PD controller through computer simulation.