An optimization method for repeated fed-batch fermentation was established with the aim of improving the recombinant human serum albumin (rHSA) production in Pichia pastoris. A simulation model for fed-batch fermentation was formulated and the optimal methanol-feeding policy calculated by dynamic programming method using five different methanol-feeding periods. The necessary state variables were collected from the calculated results and used for further optimization of repeated fed-batch fermentation. The optimal operation policy was investigated using the pre-collected state variables by estimating the overall profit per total methanol-feeding time. The calculated results indicated that the initial cell mass from the 2nd fed-batch fermentation on should be set at 35 or 40 g and methanol-feeding time at 264 h. In repeated fed-batch fermentation using the optimal operation policy, actual culture volume was in good agreement with the values simulated by model equations, but some discrepancy was observed in rHSA production. Minimum experiments were therefore carried out to re-evaluate rHSA production levels, which were then applied in re-calculations to determine the optimal operation policy. The optimal policy for repeated fed-batch fermentation established in the present study (i.e., 4-times-repeated fed-batch fermentation) achieved a 47% increase in annual rHSA production. Optimization of the culture period also brought about a 28% increase in annual rHSA production even in simple (not repeated) fed-batch fermentation. (c) 2005 Wiley Periodicals, Inc.