Abstract: A novel recursive subspace method is presented for pseudo-modal parameter identification of time-varying structures based on projection approximation subspace tracking algorithm.A first-order update matrix form suitable for on-line tracking is first derived based on matrix inversion and the characteristic of the orthogonal complement space.Then the singular value decomposition is transformed to an unconstrained optimization problem.The objective function is turned into a quadratic form of new data by projection approximation,and the eigensubspace is obtained by least square method.The pseudo-modal parameter of time-varying structure is obtained according to modal analysis theory.The pseudo-modal parameter of a manipulator with moving mass is identified and the results are compared with those of other two methods.The advantages of lower computation complexity and lower sensitivity to the noise are validated.