The time-varying mesh stiffness is a core parameter of gear systems， and the mesh stiffness identification is of great sig? nificance for the dynamic analysis and condition monitoring of gear transmission systems. Since it is difficult to directly measure the mesh stiffness， it is necessary to develop a data-driven time-varying mesh stiffness identification method. To deal with this prob? lem， an alternating state-parameter optimization method is proposed to identify the time-varying mesh stiffness of gear systems. The Fourier series with the fundamental frequency of the mesh frequency is constructed to characterize the mesh stiffness. Further? more， a Reproducing Kernel Hilbert Space （RKHS）-based de-noise method is further proposed to estimate the system state and parameter. The system state and stiffness parameter are alternately optimized with the joint constrains of dynamic model and data to realize the time-varying mesh stiffness identification of gear transmission systems. The simulation and experimental studies vali? date the effectiveness of the new mesh stiffness identification method for gear systems.