A depth guidance algorithm based on adaptive neural network is proposed to address the impact of sudden changes in seawater density on depth control during AUV underwater navigation. The algorithm designs an adaptive neural network observer to estimate the attack angle based on the AUV’s vertical plane kinematic equation, and introduces a robust term to eliminate the influence of neural network approximation errors and improve the observation accuracy of the attack angle; Then, the improvements are made to the traditional line of sight navigation method, to compensating for the command pitch angle generated by the line of sight navigation method with the observed attack angle, and enhance the depth control ability of AUV in case of sudden changes in seawater density; Finally, the Lyapunov function was designed to prove the stability of the modified derivative law. The simulation results show that the depth guidance algorithm proposed in this paper can effectively improve the depth control ability and anti-interference ability of AUV in case of sudden changes of seawater density.