The relative specificity and bond cleavage pattern of barley alpha-amylase 1 (AMY1) were dramatically changed by mutation in F(286)VD that connected beta-strand 7 of the catalytic (beta/alpha)(8)-barrel to a succeeding 3(10)-helix. This conserved tripeptide of the otherwise variable beta --> alpha segment 7 lacked direct ligand contact, but the nearby residues His290 and Asp291 participated in transition-state stabilization and catalysis. On the basis of sequences of glycoside hydrolase family 13, a biased random mutagenesis protocol was designed which encoded 174 putative F(286)VD variants of C95A-AMY1, chosen as the parent enzyme to avoid inactivating glutathionylation by the yeast host. The FVG, FGG, YVD, LLD, and FLE mutants showed 12-380 and 1.8-33% catalytic efficiency (k(cat)/K(m)) toward 2-chloro-4-nitrophenyl beta-D-maltoheptaoside and amylose DP17, respectively, and 0.5-50% activity for insoluble starch compared to that of C95A-AMY1. K(m) and k(cat) were decreased 2-9- and 1.3-83-fold, respectively, for the soluble substrates. The starch:oligosaccharide and amylose:oligosaccharide specificity ratios were 13-172 and 2.4-14 for mutants and 520 and 27 for C95A-AMY1, respectively. The FVG mutant released 4-nitrophenyl alpha-D-maltotrioside (PNPG(3)) from PNPG(5), whereas C95A-AMY1 produced PNPG and PNPG(2). The mutation thus favored interaction with the substrate aglycon part, while products from PNPG(6) reflected the fact that the mutation restored binding at subsite -6 which was lost in C95A-AMY1. The outcome of this combined irrational and rational protein engineering approach was evaluated considering structural accommodation of mutant side chains. FVG and FGG, present in the most active variants, represented novel sequences. This emphasized the worth of random mutagenesis and launched flexibility as a goal for beta --> alpha loop 7 engineering in family 13.