Abstract Nucleobase-containing coenzymes are considered the relics of an early RNA-based world that preceded the emergence of protein domains. Despite the importance of coenzyme-protein synergisms, their emergence and evolution remain poorly understood. An excellent target to address this issue is the Rossman fold, the most catalytically diverse and abundant protein architecture in Nature. Here, we investigatedted the two largest Rossman lineages, namely the nicotinamide adenine dinucleotide phosphate (NAD(P))-binding and the S-adenosyl methionine (SAM)-dependent superfamilies. With the aim to identify the evolutionary changes that lead to a switch in coenzyme specificity on these superfamilies, we performed structural and sequence-based Hidden Markov Models to systematically search for key motifs in their coenzyme-binding pockets. Our analyses revealed how insertions and deletions (InDels) reshaped the ancient β1−loop−α1 coenzyme-binding structure of NAD(P) into the well-defined SAM-binding β1−loop−α1 structure. To prove this observation experimentally, we removed an InDel of three amino acids from the NAD(P) coenzyme pocket and solved the structure of the resulting mutant, revealing the characteristic features of the SAM-binding pocket. To confirm the binding to SAM, we performed isothermal titration calorimetry measurements, validating the successful coenzyme switch. Molecular dynamics simulations also corroborated the role of InDels in abolishing NAD-binding and acquiring SAM binding. Our results uncovered how Nature utilized insertions and deletions to switch coenzyme specificity, and in turn, functionalities between these superfamilies. This work also establishes how protein structures could have been recycled through the course of evolution to adopt different coenzymes and confer different chemistries. Significance Statement Cofactors are ubiquitous molecules necessary to drive about half of the enzymatic reactions in Nature. Among them, organic cofactors (coenzymes) that contain nucleotide moieties are believed to be relics of a hypothetical RNA world. Understanding coenzyme-binding transitions sheds light onto the emergence of the first enzymes and their chemical diversity. Rossmann enzymes bind to 7 out of 10 nucleotide coenzymes, representing an ideal target to study how different coenzyme specificities emerged and evolved. Here we demonstrated how insertions and deletions reshape coenzyme-specificity in Rossmann enzymes by retracing the emergence of the SAM-binding function from an NAD-binding ancestor. This work constitutes the first example of an evolutionary bridge between redox and methylation reactions, providing a new strategy to engineer coenzyme specificity.