In a study of 523 normal subjects of differing ethnic groups, including 189 South American Indians, we have described novel hybridization patterns corresponding to 22 potentially new HLA-B locus alleles. Three of these alleles were subtypes of B35. The locally assigned alleles, B-3504v, B-3505v, and B-3508v have been sequenced and were officially designated as B*3512, B*3517, and B*3518, respectively. In addition, we determined the nucleotide sequence of another new variant, locally designated B-3509.2. B*3517, was found in 3 individuals (2 Hispanic, 1 Caucasian), it differs from B*3505 by 3 nucleotide substitutions that lead to changes in residues 94, 95, and 103. B*3517 differs from B*3501 in residues 97 and 103. B*3518 was found in 7 South American Indian individuals (6 of 124 Toba Indians, 1 of 18 Pilaga Indians). It differs from B*3509 by 2 silent nucleotide substitutions and by one nonsynonymous substitution in codon 156 (Arg → Leu). B*3512 differs from B*3504 by 3 nucleotides, one of them leading to a substitution in residue 103 (Val → Leu). B*3509 was observed in 3 individuals from the Wichi tribe. The nucleotide sequence of one of these was determined and was found to differ from B*35091 by two synonymous nucleotide substitutions. The distinguishing amino acid substitutions in residues 95, 97, and 156 contribute to the structure of specificity pockets F, C, and E, and D and E respectively; therefore, it is possible that some of the new alleles may have different peptide binding profiles. It has been shown that differences at residue 156 may elicit different allorecognition and mediate graft-versus-host disease and rejection in bone marrow transplantation. The mechanisms for the generation of these novel alleles may involve gene conversion events in which short exon-3 segments from the common Native American alleles B*4002 or B*4801 were inserted in HLA-B35 backbone structures. The novel allele B*3518 is closely related to B*35092 and to B*3508. Two alternative hypotheses for its generation can be suggested, the most plausible one would involve B*35092, the putative progenitor of B*3518, since both alleles are prevalent in the same Indian tribes.