The segregation of classical and nonclassical 21-hydroxylase deficiency (21-OHD) and its linkage to HLA-B was investigated in 220 families. First, the surprisingly high frequency of the nonclassical 21-OHD gene estimated elsewhere was confirmed using a different methodology which avoided particular assumptions concerning the classification of an individual's genotype. In the present study the gene frequency was found to be .103 +/- .020 in an ethnically pooled sample and was as high as .223 +/- .062 among Ashkenazi Jews. Second, the segregation analysis of families ascertained through a nonclassical 21-OHD proband and those ascertained through a classical 21-OHD proband showed essentially identical results. A partial recessive model with no recombination between 21-OHD and HLA-B fitted the data better than did a complete recessive model with approximately 0.5% recombination between 21-OHD and HLA-B. The support for the partial over the complete recessive model depended on the assumed ascertainment probability, an unknown parameter in these data. Four families provided most of the evidence against the complete recessive model. All these included an unaffected sib who shared both HLA-B specificities in common with the affected proband. Possible explanations for the condition in these families include recombination, gene conversion, mutation in one of the parental gametes, or technical errors.