When a disease is rare in the general population, genotype relative risks can be estimated from case-control data by adjusting the genotype frequencies in controls to the expected Hardy-Weinberg proportions. The resulting risk estimates are shown to have a smaller variance than those obtained from the observed genotype frequencies, and test statistics are given for adjusted estimates. The tests are shown to have a larger power for detecting genotype association, and heterogeneity in the genotypic relative risks, than those based on the observed frequencies. An application is given to data on rheumatoid arthritis and HLA-DR antigens.