Abstract: A direct algorithm for coefficients of generalized predictive controller is developed to simplify the analysis and design of generalized predictive control(GPC).The control law of GPC is transformed into the product of the controller coefficients,the reference trajectory and historical information of the process by recursively operating the process model.This algorithm simplifies the GPC design and reduces the on-line computational overhead because the controller coefficients are decided only by the design parameters and the process model parameters,and there is no need to compute Diophantine equations,model predictive outputs and free responses.The feasibility and efficiency of this algorithm is demonstrated by the comparison experiments on a nonlinear liquid process controlled by DCS.