The control system with the split control loop always has nonlinear characteristics, therefore, a method of split control implementation is proposed for multi-variable model predictive control. In each linear interval of the split control loop, the predictive models are established respectively according to different combination of the split valves. The control models are chosen by comparing the controller signal output values with the predefined middle threshold. After the control models being switched, the predicted output values of the old control model are regarded as the original output predicted values of the proposed model. The design problems of the multi-variable model predictive controller are solved when the process is needed to be split-range control, which enhances the applicability of the linear MPC. Finally, the simulation examples are given to verify the effectiveness of the proposed method.