A new channel estimator is proposed for Turbo receiver in MIMO-OFDM systems. The channel estimation error was taken into account in the evaluation of the extrinsic information for the coded bits. With the help of the candidates list of the list sphere decoder (LSD) and a prior information from the decoder, the approximation of the soft information in the conventional EM channel estimator was amended, and the statistic information of the soft symbols was improved for channel estimation. A new Bayesian EM channel estimation algorithm, which considers the autocorrelation of the soft symbols, was obtained. The simulation results show that the newly proposed algorithm has better performance in terms of both BER and mean square estimation error in comparison with the conventional EM algorithm.