Non-equilibrium catalysis of electron transfer reactions (i.e. redox) regulates the flux of key elements found in biological macromolecules. The enzymes responsible, oxidoreductases, contain specific transition metals in poorly sequence-conserved domains. These domains evolved ∼2.4 billion years ago in microbes and spread across the tree of life. We lack understanding of how oxidoreductases evolved; divergence of sequences makes identification difficult. We developed a method to recognise the various versions of these enzyme-domains in unannotated sequence-space. Often, homology is used to transfer function annotations from experimentally resolved domains to unannotated sequences. Unreliability of inferring homology below 30% sequence identity limits single-sequence based searches. Misaligned functional sites may compromise annotation transfer from even very similar sequences. Combining profile-based searches with knowledge of functional sites could improve domain detection accuracy. Here we present an approach that enhances the search for redox domains using catalytic site annotations. From the scientific literature, we validated annotations of 104 InterPro domains indicated as using “transition metals in redox reactions.” These domains mediate electron transfer in 20% of oxidoreductases, primarily employing iron, copper and molybdenum. We used the experimentally identified catalytic residues in these domains to validate sequence alignment-based protein function annotations. Our method, TrAnsFuSE, is 11% and 14% more accurate than PSI-BLAST and InterPro, respectively. Moreover, it is robust for use with other functional residues-we attain higher accuracy at comparable coverage using metal binding, in addition to catalytic, sites. TrAnsFuSE can be used to focus the study of the vast amounts of unannotated sequencing data from meta-/genome projects.