A new approach to subspace-based channel estimation for multiple-input-multiple-output orthogonal frequency-division multiplexing (MIMO-OFDM) systems is proposed. The proposed method exploits both the receive diversity and the statistical property of the received signal based on the iterative algorithm. To reduce the effect of residual error affecting the accuracy of the noise subspace, a measure called the Hermitian angle is introduced. In addition, the forgetting factor is used to consider a time-varying channel