D-Hydantoinase from Bacillus stearothermophilus SD1 is an important enzyme to synthesize commercially important non-natural amino acid but it has relatively low activity for aromatic hydantoin derivatives. Previously it was proven that the stereochemistry gate loops (SGLs) in the substrate binding pocket of D-Hydantoinase, (β/α)8-barrel enzyme, are major structural determinants of the substrate specificity. Based on previous studies, we attempted to computationally design the D-hydantoinase to improve specificity for hydroxyphenylhydantoin (HPH), and to provide D-p-HPG, a commercially important non-natural amino acid. Target amino acid sites were determined by experimental and structural analysis of active site of Hydantoinase. We selected Met63 and Phe159 amino acid sites on SGLs because they contact with hydroxyl ring of HPH. The result of first round was not good enough for prediction. But as the parameters of scoring function were optimized by feedback of experimental results, improved scoring function predicted another candidate sequences and we produced Hydantoinase with 4.5 times and 5.3 times higher specific activity toward HPH than wild type enzyme in second round and third round. In terms of specificity (HPH/hydantoin), improvement was 41 times and 135 times higher than wild type. Through the rational analysis of mutant activities which were designed by computational modeling, we supposed that $63^{rd}$ and $159^{th}$ amino acid residues have complementarities in amino acid side chain length. Except mutants of the highest activity, the others showed decreasing tendency when side chain lengths in $63^{rd}$ amino acid site are longer or shorter than that of the high mutants.