To address the problem of depth tracking and attitude control of autonomous underwater vehicle (AUV) near the surface, a novel nonlinear reduced-order state observer (ROSO) and a predictive controller based on nonlinear model online linearization (PC-NMOL) are presented. By using a nonsingular coordinate transformation, the ROSO is achieved to accurately estimate the state variables of AUV. And the state estimation is applied to the predictive controller to enhance the attitude control and depth tracking performance of AUV. In simulation of AUV longitudinal motion control, the comparison has been presented between ROSO and full-order state observer (FOSO), also between PC-NMOL and traditional nonlinear predictive control (NPC). Simulation results show the fast dynamical response and strong robustness of proposed methods.