Abstract Some members of the glutathione peroxidase (GPx) family have been reported to accept thioredoxin as reducing substrate. However, the selenocysteine-containing ones oxidise thioredoxin (Trx), if at all, at extremely slow rates. In contrast, the Cys homolog of Drosophila melanogaster exhibits a clear preference for Trx, the net forward rate constant, k′ +2, for reduction by Trx being 1.5 × 10 6 M − 1 s − 1 , but only 5.4 M − 1 s − 1 for glutathione. Like other CysGPxs with thioredoxin peroxidase activity, Drosophila melanogaster ( Dm)GPx oxidized by H 2O 2 contained an intra-molecular disulfide bridge between the active-site cysteine (C45; C P) and C91. Site-directed mutagenesis of C91 in DmGPx abrogated Trx peroxidase activity, but increased the rate constant for glutathione by two orders of magnitude. In contrast, a replacement of C74 by Ser or Ala only marginally affected activity and specificity of DmGPx. Furthermore, LC-MS/MS analysis of oxidized DmGPx exposed to a reduced Trx C35S mutant yielded a dead-end intermediate containing a disulfide between Trx C32 and DmGPx C91. Thus, the catalytic mechanism of DmGPx, unlike that of selenocysteine (Sec)GPxs, involves formation of an internal disulfide that is pivotal to the interaction with Trx. Hereby C91, like the analogous second cysteine in 2-cysteine peroxiredoxins, adopts the role of a “resolving” cysteine (C R). Molecular modeling and homology considerations based on 450 GPxs suggest peculiar features to determine Trx specificity: (i) a non-aligned second Cys within the fourth helix that acts as C R; (ii) deletions of the subunit interfaces typical of tetrameric GPxs leading to flexibility of the C R-containing loop. Based of these characteristics, most of the non-mammalian CysGPxs, in functional terms, are thioredoxin peroxidases.