This paper presents a computationally efficient decoder for multiple antenna systems. The proposed algorithm can be used for any constellation (QAM or PSK) and any labeling method. The decoder is based on matrix-lifting semi-definite programming (SDP). The strength of the proposed method lies in a new relaxation approach applied to the previous work by Mobasher This results in a reduction of the number of variables from (NK + 1)(NK + 2)/2, in the previous work by Mobasher to (2N +K)2 , in the new method, where N is twice the number of transmit antennas and K is the number of constellation points in each real dimension. It is shown that this reduction in the number of variables results in a significant computational complexity reduction compared to the previous work by Mobasher Moreover, the proposed method offers a better symbol error rate performance as compared to some known and recent SDP-based quasi-maximum likelihood detection methods reported in the literature.