Abstract: A novel frequency-domain subspace system identification based on forgetting factor was discussed.The key idea was applied for frequency-domain MIMO state-space models starting from frequency response function data and generalized by taking into account the initial and final conditions.It was necessary to guarantee leakage-free effect calculated through the discrete Fourier transform.A forgetting factor was introduced in the Hankel matrices of the input-output data to increase the convergent rate.In order to solve the system matrices problem,principle component analysis was used to determine the system observability subspace matrices and achieve consistent estimates of the system matrices.Before applying the algorithm,the innovation variance criterion was used to estimate the system′s order in order to decrease the complexity.The efficiency of this method was illustrated with a simulation example.