A macrokinetic model describing mycelium growth and product formation in fed-batch fermentation of Penicillium chrysogenum is proposed on the basis of intracellular metabolic pathway networks found in the literature. The model is composed of the stoichiometric balance equations for carbon source, ATP, NADH and pyruvate. A regulator model is used as an auxiliary part to simulate the lag phase at the beginning of product synthesis. Combining the macrokinetic model with the bioreactor model, one yields the relationship between manipulating variables, i.e., the substrate feeding rate, and the other state variables such as biomass, substrate and product concentrations. The rolling parameter identification is then applied to online identify the most sensitive model parameters dealing with the model mismatch caused by time-variant kinetic changes. Model validation was performed with the data of 22 batches from industrial-scale penicillin cultivation. Finally, the predictions of the product concentration, up to 24 h ahead, are carried out with an average prediction error of less than 3.5%. (C) 2010 Published by Elsevier Ltd.