A subspace based approach for the blind multiple signal separation and recovery for MIMO systems is proposed in this paper. Instead of using the statistics of the received signal, the proposed algorithm exploits the received signal structure and the finite alphabet property of the desired signals. The finite alphabet property is used to remove the unknown unitary matrix that is associated with most of the statistics-based MIMO system identification algorithms. The proposed algorithm also incorporates an error-correcting procedure; therefore, it has more accuracy than the existing algorithms. Computer simulation results demonstrate that the algorithm can detect the signals and estimate channel parameters accurately with very few symbols, even under high noise and bad channel conditions.