2002] provide convincing evidence that scoring alleles, instead of genotypes, is more powerful for family-based tests of association. Their work was motivated by prior publications that indicated that genotype tests can be more powerful than allele tests when there are only two alleles, particularly for recessive effects [Weinberg et al., 1998;Schaid, 1999]. The main reason for a decrease in power with genotype tests is the increase in degrees of freedom (df). For K alleles, the allele-based test has (KÀ1) df, which is much smaller than the genotypebased test that has K(K+1)/2À1 df, and this discrepancy gets larger as K increases. Fallin et al. [2002] concluded that when there are at least five alleles, the allele-based test is less powerful. However, inspection of their Table II shows that the allele-based test is more powerful for only 3 of their 8 simulation scenarios: scenario 3 (dominant with relative risk of 4) with K 3; scenario 4 (recessive with relative risk of 2) with K 3; and scenario 8 (recessive with relative risk of 9) for K ranging over 2-10. So, except for large risks for a recessive factor, it appears that the genotype test provides greater power than the allele test only when there are three or fewer alleles. These results are in close agreement to those presented in Figure 1 of Cleves et al. [1997]. In that report, the authors had compared the power of the test of marginal homogeneity for the transmitted vs. nontransmitted alleles vs. the test of complete symmetry. If n ij is the number parents that transmit allele i and do not transmit allele j, then the test of marginal homogeneity tests whether n i ¼ n i , which has (KÀ1) df, and the test of complete symmetry tests whether n ij ¼ n ji , which has K(KÀ1)/2 df. Hence, the test for marginal homogeneity is close to the logistic model for alleles, yet the test for complete symmetry has fewer degrees of freedom than the logistic model for genotypes. Even in this situation, the authors found that when there are more than three alleles, the test based on alleles had greater power than the test based on genotypes. So, my interpretation of the results by Fallin et al. [2002] is that the genotype test should be used only when there are 2 or 3 alleles. Furthermore, if there is evidence of a recessive factor with high risk, the genotype test would not be the n