Wireless system capacity can be increased enormously by using multiple antennas at transmitter and receiver sides in the system. Among them, maximum likelihood (ML) receiver is an optimal receiver, but it is difficult to implement due to high complexity arising from exhaustive searches over all candidate vector symbols. On the other hand, zero forcing ordered successive interference cancellation (ZF-OSIC) receiver can reduce the complexity greatly although its performance is less than ML receiver. Performance of ZF-OSIC receiver is seriously affected by the lower SNR layers. Therefore, in our proposed detection scheme, we will choose the optimal number of transmit antennas in the high SNR transmit layers, to be detected by the ZF-OSIC scheme and the remaining transmit antennas in the low SNR transmit layers will be detected by ML scheme to give the better BER performance for the spatial multiplexing MIMO system. Simulated results show that the proposed scheme gives better system performances than those of the conventional ZF-OSIC scheme not only in complexity but also in BER condition.