Recent reports describe a new strategy for the binding of single-stranded pyrimidine sequences by triple helix formation. In this approach, a double-length purine-rich oligonucleotide binds a target strand, folding back to form an antiparallel pur.pur.pyr triple helix. We now describe a series of studies in which sequence and structural variations are made in such purine-rich ligands, in an effort to optimize binding properties. Comparison is made between the use of two separate strands and the use of single two-domain ligands; the latter are found to bind more tightly and to aggregate less in media containing Na+ or K+. Placement of mismatched bases in the target shows that sequence selectivity of binding is as high as that for Watson-Crick duplex formation. Variation of the lengths and sequences of loops bridging the binding domains demonstrates that dinucleotide loops composed of pyrimidines give the highest stability. Oligoethylene glycol-derived loop replacements are shown to give good binding affinity as well. The binding of an RNA target is shown to occur with the same affinity as the binding of DNA. In general, it is found that circular variants bind more tightly than do either separate strands or singly-linked ligands and unlike linear oligomers, the circular compounds do not aggregate to a large extent even in buffers containing 100 mM K+. Such structurally optimized ligands are useful in expanding the number of possible naturally-occurring sequences which can be targeted by triplex formation.