Because the aspartic acid (Asp) residues in proteins are occasionally isomerized in the human body, not only l-α-Asp but also l-β-Asp, d-α-Asp and d-β-Asp are found in human proteins. In these isomerized aspartic acids, the proportion of d-β-Asp is the largest and the proportions of l-β-Asp and d-α-Asp found in human proteins are comparatively small. To explain the proportions of aspartic acid isomers, the possibility of an enzyme able to repair l-β-Asp and d-α-Asp is frequently considered. The protein l-isoaspartyl (d-aspartyl) O-methyltransferase (PIMT) is considered one of the possible repair enzymes for l-β-Asp and d-α-Asp. Human PIMT is an enzyme that recognizes both l-β-Asp and d-α-Asp, and catalyzes the methylation of their side chains. In this study, the binding modes between PIMT and peptide substrates containing l-β-Asp or d-α-Asp residues were investigated using computational protein–ligand docking and molecular dynamics simulations. The results indicate that carboxyl groups of both l-β-Asp and d-α-Asp were recognized in similar modes by PIMT and that the C-terminal regions of substrate peptides were located in similar positions on PIMT for both the l-β-Asp and d-α-Asp peptides. In contrast, for peptides containing l-α-Asp or d-β-Asp residues, which are not substrates of PIMT, the computationally constructed binding modes between PIMT and peptides greatly differed from those between PIMT and substrates. In the nonsubstrate peptides, not inter- but intra-molecular hydrogen bonds were observed, and the conformations of peptides were more rigid than those of substrates. Thus, the in silico analytical methods were able to distinguish substrates from nonsubstrates and the computational methods are expected to complement experimental analytical methods.