The covariance matrix algorithms have strong robustness under the condition of large mismatch errors of MIMO radar.Howeverthose algorithms usually have high computational complexity.In view of the contradiction between high performance and low computational complexity, this paper proposes a method to modify the mismatch error model. The model is modified from two aspects. Firstly, based on bilateral constraining, the error set of the unilateral steering vector is set up to constraint the errors of the transmitter and the receiverand the mismatch error set of the joint steering vector was obtained. Secondly, based on the structural characteristics of the uncertainty set model, K-means clustering algorithm is used to obtain a small number of weighted characteristic sampling points to replace the previous large number of discrete sampling points, so as to reduce the number of sampling points needed for covariance matrix construction and thus reduce computational complexity. Simulation results show that the proposed algorithm has strong robustness under the condition of large mismatch errors and presents outstanding performance in improving the output SINR.