Industrial fermentation processes typically use complex media and operate in fed-batch mode to minimize the effects of catabolite repression. However, model-based feeding recipes have not been reported for such processes primarily as a result of the lack of reliable process models. By using a recently published process model, we optimize the feeding recipe for rifamycin B fermentation in complex media. Experimental validation shows a twofold improvement in productivity over an optimized batch. The dynamic optimization problem was posed as a nonlinear program and solved using successive quadratic programming. The feed profiles of four substrates were parameterized to convert the problem into a finite decision space consisting of substrate feed rates and switching intervals. Several distinct recipes, each corresponding to a unique initial guess of decision variables, showed comparable productivity, implying the presence of multiple local optima. The strategy presented here can be applied for optimization of other fermentation processes for which reliable process models are available. (c) 2006 American Institute of Chemical Engineers.