Résumé : The 12 genes which in E. coli K-12 constitute the arginine regulon are organized in nine transcriptional units all of which contain in their 5′ non-coding region two 18 bp partially conserved imperfect palindromes (ARG boxes) which are the target sites for binding of the represser, a hexameric protein. In vitro binding experiments with purified repressor (a gift from W. K. Maas) were performed on the operator sites of four genes, argA, argD, argF, argG, and of two operons, carAB and the bipolar argECBH cluster. A compilation of results obtained by DNase I and hydroxyl radical footprinting clearly indicates that in each case the repressor binds symmetrically to four helical turns covering adjacent pairs of boxes separated by 3 bp, but to one face of the DNA only. Methylation protection experiments bring to light major base contacts with four highly conserved G residues symmetrically distributed in four consecutive major grooves. Symmetrical contacts in the minor groove with A residues have also been identified. Stoichiometry experiments suggest that a single hexameric repressor molecule binds to a pair of adjacent ARG boxes. Although the wildtype operator consists of a pair of adjacent ARG boxes separated by 3 bp (except argR where there are only 2 bp), repressor can bind to a single box but with a greatly reduced affinity. Therefore, adjacent boxes behave co-operatively with respect to the Arg repressor binding, in the sense that the presence of one box largely stimulates the binding of the properly located second box. The optimal distance separating two boxes is 3 bp, but one bp more or less does not abolish this stimulation effect. However, it is completely abolished by the introduction of two or more additional bp unless a full helical turn is introduced. Large variations in the in vivo repression response between individual arginine genes or a wild-type gene and cognate Oc type mutants are not reflected by similar differences in the in vitro binding results where only small differences are observed. The significance of this lack of correlation is discussed. © 1992.