An on-line adaptive optimization algorithm based on input-output model identification was, in the present simulation study, applied to a high-density cell culture of Saccharomyces cerevisiae for ethanol production from glucose. The model system consisted of a continuous bioreactor, a membrane filter unit for cell recycle, and pumps for transport. The objective of optimization was to maximize the ethanol productiv-ity by manipulating dilution rate at a fixed bleeding ratio. The bilevel forgetting factor(BFF) method was used for the estimation of parameters of the input-output model which was employed for on-line system identification. during the coures of optimization new values of the dilution rate were recurrently determined by using the steepest ascent method based on results of the model parameter estimation. When no ethanol measurement noise was involved the system was driven and maintained closely to the optimum point in about 250 hours by using the proposed algorithm. when the feed glucose concentration was changed while the culture was maintained at the optimum, the proposed algorithm could recognize the change and drove the culture to a new optimum point. The algorithm also showed a good operational stability which could stably maintain the culture at its optimum for about 3000 hours. Even when the measured concentrations were influenced by noises, the average value of the ethanol productivity was maintained closely to its optimum although the overall performance was relatively unstable and the dilution rate showed significant fluctuations.