The traditional model updating method based on frequency response function （FRF） is difficult to converge in case of large test noise， large residual between initial analysis FRF and test FRF， or large number of parameters to be modified. A dynamic finite element model （FEM） updating method based on maximum likelihood estimation and frequency shift is presented. The maximum likelihood estimator using "a priori" variance information of the FRF is constructed to obtain the optimal parameter estimation depending on the Gauss-Newton method. The frequency shift method and the total least square adjustment method are used to solve the iterative equation to improve the robustness of parameter identification. The test data of poor quality is eliminated according to selection criterion of frequency points. An approach for expanding incomplete experimental FRF is presented to reduce the extra error. Finally， numerical simulation and experimental investigation for updating the dynamic finite element model of a bogie structure and an aircraft model are performed to validate the robustness and efficiency of the present method. The obtained results show that the present method can be successfully applied to update the complex finite element models.