Abstract: This paper researched on the optimization of multivarate exponentially weighted moving average(MEWMA) control chart. In the retrospective phase, a new covariance matrix S2 based on successive differences is recommended for estimating the common covariance matrix Σ. An improved MEWMA chart based on estimator S2 was established. When step shifts or ramp shifts in mean value occur, the signal probabilities and ARLs(average run length) of two MEWMA charts are compared. According to the simulation results, the new control chart presents a remarkable optimization with large smoothing factor r and noncentrality parameter λ.