Markers in the 3' end of the vitamin D receptor gene have recently been associated with prostate cancer risk. To evaluate the adequacy of the commonly used BsmI restriction fragment length polymorphism as a marker of this locus, we genotyped 627 individuals from five ethnic groups for this marker, as well as for a polymorphic site in the 3' untranslated region of this gene. At the latter site, we identified 12 alleles, A13 to A24, of a poly(A) microsatellite. Allele size followed a bimodal distribution with distinct short (A13-A17) and long (A18-A24) allele populations. Poly(A) allele frequency differed by ethnicity, with the frequency of short alleles being highest in non-Hispanic whites (41%), intermediate in Hispanics and African-Americans (31 and 29%, respectively), and lowest in Japanese-Americans and Chinese (8 and 9%, respectively). In each of the ethnic groups, some degree of coupling was observed between BsmI B and short poly(A) alleles and between BsmI b and long poly(A) alleles. However, the strength of the linkage disequilibrium varied by ethnicity, with departures from complete disequilibrium producing disagreement between the BsmI and poly(A) genotypes. Genotypic disagreement was lowest in Japanese-Americans and non-Hispanic whites (6 and 7%, respectively), intermediate in Chinese and Hispanics (11 and 19%, respectively), and highest among African-Americans (37%), indicating that BsmI is not a good marker for the vitamin D receptor 3' untranslated region genotype in all populations. This finding may explain contradictory results from recent association studies using the BsmI marker.