Enzymatic stereodivergent synthesis to access all possible product stereoisomers bearing multiple stereocenters is relatively undeveloped, although enzymes are being increasingly used in both academic and industrial areas. When two stereocenters and thus four stereoisomeric products are involved, obtaining stereodivergent enzyme mutants for individually accessing all four stereoisomers would be ideal. Although significant success has been achieved in directed evolution of enzymes in general, stereodivergent engineering of one enzyme into four highly stereocomplementary variants for obtaining the full complement of stereoisomers bearing multiple stereocenters remains a challenge. Using Candida antarctica lipase B (CALB) as a model, we report the protein engineering of this enzyme into four highly stereocomplementary variants needed for obtaining all four stereoisomers in transesterification reactions between racemic acids and racemic alcohols in organic solvents. By generating and screening less than 25 variants of each isomer, we achieved >90% selectivity for all of the four possible stereoisomers in the model reaction. This difficult feat was accomplished by developing a strategy dubbed "focused rational iterative site-specific mutagenesis" (FRISM) at sites lining the enzyme's binding pocket. The accumulation of single mutations by iterative site-specific mutagenesis using a restricted set of rationally chosen amino acids allows the formation of ultrasmall mutant libraries requiring minimal screening for stereoselectivity. The crystal structure of all stereodivergent CALB variants, flanked by MD simulations, uncovered the source of selectivity.