A full understanding of the molecular basis for genetically determined human traits, including susceptibility to disease, appears to be within reach following recent breakthroughs. How fully this promise will be realized, and by which combination of study designs, will depend to a large extent on the allelic architecture of each trait, which is still unknown in most cases. The prevailing belief that traits common in the general population must depend on common variants is challenged by theoretical predictions based on the mutation-selection model. This model states that if disease variants are subject to even weak purifying selection, their presence can be maintained only by new mutations, resulting in a multitude of rare alleles at each locus. Predictions favoring each scenario have relied on biased evidence and unverifiable assumptions, respectively. However, unbiased factual testing of them may soon be possible, as data accumulate from genome-wide association studies and high-throughput resequencing. Because the models are not mutually exclusive, the question should be not which model is correct, but rather what is the relative contribution of each, which is something that may vary dramatically among traits.