Abstract: As one of the important intelligent modeling methods, BP neural network can be effectively used for GSH fermentation process modeling. However, it usually lacks the nicer interpretation and its performance is often deteriorated by noise existing in the sample data. In order to circumvent theses weaknesses, a robust and interpretable syllogistic-fuzzy-inference based modeling method, called CCTSK (Cascade Centralized TSK) fuzzy neural network, is introduced for GSH fermentation process modeling. Present experimental results demonstrate that the proposed method really has better robustness compared with the traditional BP neural network.