This paper is concerned with the production of copolymers with uniform copolymer composition and desired weight-average molecular weight by using a learning-based nonlinear model predictive control (NLMPC). For an effective control in a semibatch copolymerization system, the successive linearization method used in NLMPC is modified. The nonlinear model for the copolymerization system is linearized by using the previous batch data within the prediction horizon in such a way that the linear time-varying model is obtained as a function of the increment of the inputs between the two consecutive batches. The learning algorithm is incorporated into the controller by virtue of the model structure, and the effectiveness of the proposed controller is shown by comparison with a conventional nonlinear model predictive controller as well as a linear model-based learning controller. Through the implementation of the controller to a semibatch methyl methacrylate/methyl acrylate copolymerization reactor, it is proven that copolymers with desired properties are produced effectively using the algorithm proposed in this study.