An industrial fed batch fermentation process is a nonlinear time-varying process. Important internal state variables such as biomass, substrate and product concentrations cannot be measured online and are usually determined by infrequent and time consuming off-line laboratory analysis. The online measurements are usually noisy and sometimes this leads to misinterpretation of the real situation inside the fermenter. These problems can lead to poor control of the batch and low productivity subsequently. To overcome these problems a real time expert system has been proposed which is based on the Poplog Flex real time expert system shell. The system is used to monitor the state variables of the process, diagnose any fault that might occur in the process, estimate the important unmeasurable state variables and to design a controller to control the state around a desired level. A neural network has been adopted for the online estimation of the unmeasurable state variables. Pattern recognition ideas have been used to improve the modelling ability of the neural network. Predictive control techniques have been used to control the state around a desired level. The model and the controller for the process have been designed and implemented within the Poplog Flex environment.<<ETX>>