ADP-glucose pyrophosphorylase (AGPase) catalyzes a rate-limiting step in glycogen and starch synthesis in bacteria and plants, respectively. Plant AGPase consists of two large and two small subunits that were derived by gene duplication. AGPase large subunits have functionally diverged, leading to different kinetic and allosteric properties. Amino acid changes that could account for these differences were identified previously by evolutionary analysis. In this study, these large subunit residues were mapped onto a modeled structure of the maize (Zea mays) endosperm enzyme. Surprisingly, of 29 amino acids identified via evolutionary considerations, 17 were located at subunit interfaces. Fourteen of the 29 amino acids were mutagenized in the maize endosperm large subunit (SHRUNKEN-2 [SH2]), and resulting variants were expressed in Escherichia coli with the maize endosperm small subunit (BT2). Comparisons of the amount of glycogen produced in E. coli, and the kinetic and allosteric properties of the variants with wild-type SH2/BT2, indicate that 11 variants differ from the wild type in enzyme properties or in vivo glycogen level. More interestingly, six of nine residues located at subunit interfaces exhibit altered allosteric properties. These results indicate that the interfaces between the large and small subunits are important for the allosteric properties of AGPase, and changes at these interfaces contribute to AGPase functional specialization. Our results also demonstrate that evolutionary analysis can greatly facilitate enzyme structure-function analyses.