Although studies suggest that SNPs derived from HapMap provide promising coverage and power for association studies, the lack of alternative variation datasets limits independent analysis. Using near-complete variation data for 76 genes resequenced in HapMap samples, we find that coverage of common variation by commercial genotyping arrays is substantially lower compared to the HapMap-based estimates. We quantify the power offered by these arrays for a range of disease models.