The authors describe a computational approach to design enzymes capable of catalyzing specific chemical reactions, focusing on retro-aldolases that break carbon-carbon bonds in nonnatural substrates. Using new algorithms that rely on hashing techniques, they designed 72 retro-aldolases with four different catalytic motifs. Among these, 32 designs showed detectable retro-aldolase activity, with designs involving explicit water molecules for proton shuffling being significantly more effective, achieving rate enhancements of up to four orders of magnitude. The atomic accuracy of the designs was confirmed by x-ray crystal structures of active designs embedded in two protein scaffolds, which closely matched the design models. The study highlights the potential of computational design to create enzymes with novel catalytic activities and suggests that future work should focus on improving the catalytic efficiency of designed enzymes to match that of naturally occurring enzymes.The authors describe a computational approach to design enzymes capable of catalyzing specific chemical reactions, focusing on retro-aldolases that break carbon-carbon bonds in nonnatural substrates. Using new algorithms that rely on hashing techniques, they designed 72 retro-aldolases with four different catalytic motifs. Among these, 32 designs showed detectable retro-aldolase activity, with designs involving explicit water molecules for proton shuffling being significantly more effective, achieving rate enhancements of up to four orders of magnitude. The atomic accuracy of the designs was confirmed by x-ray crystal structures of active designs embedded in two protein scaffolds, which closely matched the design models. The study highlights the potential of computational design to create enzymes with novel catalytic activities and suggests that future work should focus on improving the catalytic efficiency of designed enzymes to match that of naturally occurring enzymes.