Abstract: In order to solve the remote bearing estimation problem in shallow water, a new robust bearing estimation method based on the normal mode theory is proposed in this paper. On the basis of the multi-path effects, we utilize the normal mode theory to establish the array signal model, and then impose constraint conditions on the actual source vectors to improve the robustness for the high-resolution bearing estimation methods under the model parameter uncertainties. Through the regularization method, we also present a detailed approach to convert the proposed robust bearing estimation method into the Second Order Cone Programming (SOCP) and then solve the problem by the mathematic software tool, Sedumi. At the same time, the spatial spectrum structure, Peak to Sidelobe Ratio and -3dB beamwidth are discussed and analyzed in detail. The simulation results show that the proposed algorithm has higher spatial spectrum peaks and lower sidelobe level under the conditions of different water depths and sound velocity gradient uncertainties, that is, the proposed bearing estimation method in shallow water has robust performance.