In an industrial gas-phase polymerization process, process dynamics change considerably due to grade and load changes, and these changes present a difficult control problem. In this paper, a nonlinear physical model is effectively used for control system design. First, we developed a set of physical models of an industrial ethylene polymerization process, referring to McAuley's model. Parameters in the model were adjusted in order to simulate the actual process behavior. We examine the changes of process dynamics under typical operating points for different grade products. Second, in the control system design, the process is regarded as a two-input and two-output system. The two inputs are the feed rates of fresh hydrogen and butene, and the two outputs are cumulative melt index and density. An optimal servo controller with integral actions is designed according to the optimal regulator theory using a model linearized at a nominal operating point of the target grade. As a result, the optimal servo controller with integral actions results in better control performances during critical grade changes than the PID controller does.