This paper focuses on the semi-blind equalization for Multiple-Input Multiple-Output (MIMO) systems with Quadrature Amplitude Modulation (QAM) signal. Combining the advantages of simplicity of Square Contour Algorithm (SCA) and accuracy of Soft Decision-Directed (SDD) scheme together, the SCA added SDD (SCA+ SDD) method is proposed, which has the good performances of both simplicity and accuracy. In the optimization procedure, a minimum number (equal to the number of receiving antennas) of training symbols are firstly utilized to derive the rough estimate of the spatial equalizers weight vectors, and then the conjugate gradient algorithm is proposed to optimize the cost function. The new scheme possesses the performance of approximate quadratic convergent. Compared with traditional gradient-type algorithms, conjugate gradient algorithm has a faster convergent speed and less computational quantity complexity. Finally, the methods reliability and validity are evaluated by Bite Error Rate (BER) and convergent speed respectively. Computer simulation confirms the good performances of the algorithm.