Abstract: The design of generalized predictive control (GPC) algorithm in a liquid level control system is studied, the CARIMA model of object is constructed, and the parameters selection of optimization measures in GPC is given. In order to solve the problems that are produced by the basic least squares algorithm in estimating the model parameters, the improved methods are given in the parameters recursive estimation, estimation convergence speed, estimation online verifying and in dealing with the input and output data of the object. Experimental results show that the improved methods are reasonable and practicable.