A method is described to reveal the relative predispositional effects (RPEs) (predisposing, protective, or neutral) of the HLA alleles or of any other marker system that is associated with a disease. When the disease is associated with two or more alleles of a locus, the RPE method identifies the associations sequentially according to their strength; thus the problem that a strong association with one allele can create misleading deviations in the frequencies of other alleles is alleviated. Using this method, we have examined the relative effects of HLA-DR alleles in susceptibility to Graves disease in the Caucasian population. The well-established positive association with DR3 was confirmed as the strongest effect. In addition, a negative association was found between DR5 and Graves disease. The reduced frequency of DR5 among patients is statistically significant and is not a result of the increase in DR3. Finally, when patients were divided according to the presence or absence of eye disease, the latter showed a significant increase in the frequency of DR4. With family data, linkage to HLA of Graves disease was established in both Caucasian and Chinese families by the sib-pair method.