Adaptive-network-based Fuzzy Inference System (ANFIS), proposed by Jang, is applied to estimating characteristics of end products for a semibatch process of polyvinyl acetate. In modeling the process, it is found that an ANFIS model restructured in a way of cascade mode enhances predictive performance. And membership functions for temperature, solvent fraction, initiator concentration and monomer conversion, which are changed by training, are analyzed. Consequently, it is considered that the analysis of parameter adjustment in the membership functions can clarify effect of adding the conversion to an input variable of fuzzy sets on enhancement of robustness and improvement of local prediction accuracy in restructuring ANFIS model.