Aimed at the difficulty in identifying the equivalent stiffness of rotor bearing joint, an equivalent stiffness identification method of combining finite element modal analysis, experimental modal analysis and back propagation neural network was proposed. Firstly, the finite element model analysis of a rotor bearing system was built with the Abaqus, and the respective natural frequencies were obtained. Secondly, the first four bending natural frequencies of the rotor bearing system were acquired with experimental modal analysis. Finally, the equivalent stiffness and natural frequencies were trained and identified by BP neural network. Research results showed that the maximum relative error of the radial and axial equivalent stiffness identification of the rotor bearing system were 1.75% and 5.43% respectively. The research work provides a reference for parametric identification on mechanical structure joint.