To the Editor: The personal genetics revolution has promised patients an accounting of their individual risk of common, complex diseases based on their DNA sequence. Though under increased scrutiny from the Food and Drug Administration, several direct-to-consumer (DTC) genetic testing companies offer such services, and conflicting results for the same disease in the same individual are commonly reported1. Even for an unusual case like age-related macular degeneration (AMD) for which a small number of loci with strong effects has consistently replicated across studies, it is extremely difficult to predict who will or will not develop disease2. Furthermore, most genetic association studies have been conducted in European Americans, and because the frequency of genetic polymorphisms varies across race-ethnicities, the predictive value of any genetic algorithm developed in one population may not translate to another. We have seen an extreme example of this for the ARMS2 AMD susceptibility locus. The non-synonymous coding variant A69S within ARMS2 is one of the strongest genetic risk factors for AMD (odds ratios (OR) ~2.2 in heterozygotes, ~7.1 in homozygotes3 in European Americans). This variant (or others in strong linkage disequilibrium with it) has been used in predictive algorithms published in the scientific literature2, 4, marketed by DTC companies, and in the Macula Risk™ test available by physician order.