Almost half of all enzymes utilize a metal cofactor. However, the features that dictate the metal utilized by metalloenzymes are poorly understood, limiting our ability to manipulate these enzymes for industrial and health-associated applications. The ubiquitous iron/manganese superoxide dismutase (SOD) family exemplifies this deficit, as the specific metal used by any family member cannot be predicted. Biochemical, structural and paramagnetic analysis of two evolutionarily related SODs with different metal specificity produced by the pathogenic bacterium Staphylococcus aureus identifies two positions that control metal specificity. These residues make no direct contacts with the metal-coordinating ligands but control the metal's redox properties, demonstrating that subtle architectural changes can dramatically alter metal utilization. Introducing these mutations into S. aureus alters the ability of the bacterium to resist superoxide stress when metal starved by the host, revealing that small changes in metal-dependent activity can drive the evolution of metalloenzymes with new cofactor specificity.