The crystal structure of bovine erythrocyte glutathione peroxidase has been refined by a combined procedure of restrained crystallographic refinement and energy minimization at 0.20 nm resolution. The final R value at this resolution is 0.178. The r.m.s. deviation of main-chain atoms of the two independently refined monomers is 0.019 nm. The structure at 0.28 nm resolution, which has been determined by multiple isomorphous replacement, served as a starting model.The refined model allowed a detailed survey of the hydrogen-bonding pattern and of the subunit contact areas in the molecule. The model contains 165 solvent molecules per dimer, all taken as water molecules. The mobility of the structure was derived from the individual atomic temperature factors. The complete tetramer, including the active sites, seems to be rather rigid, except for narrow loops near to the N-terminal ends and some p turns exposed to solvent.The active centres of glutathione peroxidase are found in flat depressions on the molecular surface. The catalytically active selenocysteine residues could be located at the N-terminal ends of a helices forming /YG$ substructures together with two adjacent parallel p strands. In the vicinity of the reactive group some aromatic amino acid side-chains could be localized. Especially Trp-148, which could be hydrogen bonded to SeCys-35, may play a functional role during catalysis.The results ofsubstrate and inhibitor binding studies in solution and in the crystalline state could be interpreted by an apparent half-site reactivity of glutathione peroxidase. The enzyme seems to react in the sense of negative cooperativity with dimers being the functional units.Based on difference Fourier analyses of appropriate derivatives a reasonable model of glutathione binding is presented. Among the residues which could be of functional importance are Arg-40, Gln-130 and Arg-167, presumably forming salt bridges and a hydrogen bond to the glutathione molecule.In conclusion, a general picture of a minimal reaction mechanism, which is in good agreement with functional and structural data, is proposed. The main reaction of the catalytic cycle presumably shuttles between the selenolate and the selenenic acid state of SeCys-35.