This paper investigates the trajectory tracking control problem for an autonomous underwater vehicle (AUV) subject to the external disturbance, modelling uncertainty and an initial tracking error. A trajectory tracking control strategy is proposed based on a virtual closed-loop system. The proposed control architecture is based on a virtual closed-loop system used to generate the virtual reference trajectory that the AUV should follow instead of the original desired trajectory. At first, in the virtual closed-loop system, a fixed feedback controller is designed based on an approximation dynamic model of the AUV to relieve the chattering phenomenon of control input at the early stage. Then, a robust controller based on a generalised saturation function is combined with neural network estimation to allow the AUV to track the generated virtual reference trajectory. Finally, simulations with different trajectories are performed on an AUV to evaluate the performance of the new design.