In order to avoid grinding chatter, a new method to predict grinding chatter is presented in the grinding process based on complexity of empirical mode decomposition and discrimination information. In this method, firstly, empirical mode decomposition is used to obtain the intrinsic mode function of filtered vibration signal detected in the grinding process. Then, complexity of intrinsic mode function is calculated by the L-Z complexity, and the characteristic value of grinding chatter prediction is obtained. Finally, the analysis of discriminat ioninformation is used to quantify the changing of complexity, and discrimination information is used to predict grinding chatter in grinding process. In this paper, cylindrical grinding experimental platform is used to verify the validity of this method. Three different processing which are varying workpiece speed, varying grinding wheel speed and varying grinding depth are operated to tend to chatter state, then we use this method to analyze vibration signals detected in three different processing. Experimental results show that when three different processing tend to chatter state, discrimination information will increase gradually, therefore, the value of the identification information can predict chatter in the grinding.