We describe the synthesis and the incorporation into oligonucleotides of the novel nucleoside building blocks 9, 10, and 16, carrying purine-like double H-bond-acceptor bases. These base-modified nucleosides were conceived to recognize selectively a cytosine ¥ guanine (C ¥ G) inversion site within a homopurine ¥ homopyrimidine DNA duplex, when constituent of a DNA third strand designed to bind in the parallel binding motif. While building block 16 turned out to be incompatible with standard oligonucleotide-synthesis conditions, UV/ triplex melting experiments with third-strand 15-mers containing b-d-nucleoside 6 (from 9) showed that recognition of the four natural Watson-Crick base pairs follows the order G ¥ C % C ¥ G > A ¥ T > T ¥ A. The recognition is sequence-context sensitive, and G ¥ C or C ¥ G recognition does not involve protonated species of b-d-nucleoside 6. The data obtained fit (but do not prove) a structural model for C ¥ G recognition via one conventional and one CÀH ¥¥¥ O H-bond. The unexpected G ¥ C recognition is best explained by third-strand base intercalation. A comparison of the triplex binding properties of these new bases with those of 4deoxothymine (5-methylpyrimidine-2(1H)-one, 4 H T), previously shown to be C ¥ G selective but energetically weak, is also described.