A suboptimal minimum mean-squared error estimation (MMSE) is proposed for a dispersive wireless channel in the absence of its correlation matrix for multipleinput multiple-output orthogonal frequency division multiplexing (MIMO- OFDM) transmission. It utilizes a fast subspace approximation tracking to separate signal subspace With a limited set of channel estimates. The subspace rank is adjusted by pre-set thresholds in different signal-to-noise ratios (SNRs). The performance comparison among the proposed algorithm, least square based, and the optimal MMSE estimation is shown by numerical simulation trader a spatially correlated multi-tap channel scenario. It demonstrates that the approach has better normalized mean square error than recursive least square estimation and yields 3 dB gain over the latter.