An expectation maximization (EM) algorithm for joint channel tracking and symbol detection in a multi-input multi-output (MIMO) time-varying frequency-selective fading environment is proposed in this research. Based on the recursive EM procedure in conjunction with soft decoding, we develop an iterative algorithm that performs the minimum mean squared error (MMSE) channel estimation and the maximum a posterior (MAP) probability symbol detection jointly. Two soft decoders are examined; namely, the BCJR algorithm and the soft sphere decoder. The performance of the proposed algorithm is evaluated via simulation and compared with that of Kalman filtering with hard decision feedback. It is demonstrated by numerical simulation that the proposed algorithm has robust performance in the presence of a severe channel model mismatch.