Abstract: The sensitivity-based finite element model updated with classical Tikhonov regularization can alleviate the ill-conditioning in solving the damage identification problems, and suppress the influence of noise in the measured model parameters. However the introduction of Tikhonov regularization may lead to an over-smooth solution. In order to improve the identification of non-smooth solution, smooth function is introduced in Tikhonov punishment function to enhance robustness and accuracy of the structure damage identification algorithm. The numerical simulations show that sensitivity-based model updated with improved Tikhonov regularization can reduce the influence of measurement noise effectively and identify the structure damages correctly.