A sequential optimization strategy, based on statistical experimental designs, was used to enhance the production of riboflavin by recombinant Bacillus subtilis RH44. In the first instance, the medium components were optimized in shake flask cultures. After preliminary experiments of nitrogen source selection, the two-level Plackett-Burman (PB) design was implemented to screen medium components that significantly influence riboflavin production. Among the 15 variables tested, glucose, NaNO3, K2HPO4, ZnSO4, and MnCl2 were identified as the most significant factors (confidence levels above 95%) for riboflavin production. The optimal values of these five variables were determined by response surface methodology (RSM) based on the central composite design (CCD). The validity of the model developed was verified, and the optimum medium led to a maximum riboflavin concentration of 6.65 g/l, which was 44.3 and 76.4% higher than the improved medium and the basal medium, respectively. A glucose-limited fed-batch culture profile in a 5-l fermentor was consequently designed according to the above optimum medium in shake flasks. A final riboflavin concentration of 16.36 g/l was obtained in 48 h, which further verified the practicability of this optimum trategy.