A backpropagation neural network (BPN) was applied for the control study of 2,3-butanediol fermentation (2,3-BDL) carried by Klebsiella oxytoca. The measurements of cell mass and glucose were not included in the network models, instead, only the on-line measured product concentrations from the MIMS (membrane introduction mass spectrometer) were involved. Oxygen composition was chosen to be the control variable for this fermentation system for the formation of 2,3-BDL is regulated by oxygen. Oxygen composition was directly correlated to the measured product concentrations. A twodimensional (number of input nodes by number of data sets) moving window to supply data for on-line, dynamic learning of this fermentation system was applied. The input nodes of the networks were also properly selected. Two neural network control schemes for this 2,3-BDL fermentation were discussed and compared in this work. Fermentations often exist time delay due to the measurement and their slow reaction nature. Hence, the order of time delay for the network controller was also investigated.1 Introduction 2,3-BDL is the major product from the fermentation by Klebsiella oxytoca. As a secondary metabolite, it can be produced at an optimal yield under a partially anaerobic condition [1]. Some other metabolites such as acetate, ethanol, and acetoin are also secreted as the end products from this fermentation. The major metabolic pathway of this fermentation was shown in our previous work [2]. The fermentation products shown in that work are only limited to those small molecules that secrete extracellularly. The distribution of product formation in 2,3-BDL fermentation can be governed by the amount of supplied oxygen. Therefore, the secreted product concentrations can be directly correlated to the oxygen supply. The on-line measurement of four products, acetic acid, acetoin, ethanol, and 2,3-BDL, was successfully performed by a mass spectrometer modi®ed by the insertion of a membrane probe [3]. In general, glucose and cell mass are two important measurements involved in fermentation related studies. However, these two measurements will not be referred to in this study. The dissolved oxygen (DO) in the fermentation broth re¯ects the oxygen availability to the cells. Therefore, the DO level of the fermentation can be regulated by aeration rate or oxygen composition. In this study, oxygen composition is used as the control input variable.Instead of a static model, dynamic identi®cation with adaptively adjusted parameters was applied to the control system of this fermentation. The control system is performed by the process model and the controller model. The process model is obtained from system identi®cation which was proceeded dynamically in this work. Therefore, a precise process model (or system identi®cation) and a good controller model (control law) are two major components for a control system. In general, the model from system identi®cation is combined with controller model to become a control system so that the future control input comma...