Abstract: An improved linear minimum mean square error(MMSE) blind space-time multiuser detection (MUD) algorithm and its adaptive implementation based on subspace tracking is presented for the multiple-input multiple-output Direct-Sequence Code-Division Multiple Access (MIMO DS-CDMA) communication system. The proposed detection algorithm can be implemented by three steps for improvement. Firstly, the signal subspace can be pretreated by modifying a rough estimated signal subspace with MIMO wireless channels which gained by making full use of the available information of spatial diversity technique and space-time block coding (STBC) scheme of MIMO system. Then the normalized orthogonal Oja (NOOja) algorithm with better orthogonality and stability than other algorithms is used to track the signal subspace. Aimed at the problem of detecting performance degradation on traditional algorithm caused by eigenvalue matrix approximate evaluation in every iteration, eigenvalue matrix is optimized to remove noise in the adaptive process. Simulation results demonstrate that this algorithm is efficient to suppress multi-access interference (MAI) and combat the near-far resistant. In particular, it has good tracking ability and steady-state performance in the low SNR and distinct near-far resistant environment.