The public SNP databases are an important resource for groups performing genetic association and linkage studies. Both academic and commercial groups are developing large numbers of genotyping assays for SNPs in candidate genes or spread across the genome. These databases now contain in excess of 6 million SNPs that have been generated using a large number of methods and cohorts. Today, however, only a small fraction of these SNPs are well characterized and validated. The latest release of dbSNP contains approximately 3.7 million non-redundant entries, only 0.5 million of which are validated, and 0.2 million of which have frequency information. Users of these databases have several common questions. How many of the SNPs are real? What is the frequency spectrum of the SNPs in these databases? What is the distribution picture of these SNPs across different ethnic and geographical populations? What fraction of the total number of SNPs is already captured by these databases? In order to address these questions, we compared the public SNPs against a well-characterized collection of gene-centric SNPs that we have developed. From this comparison, we find that > 50% of high frequency SNPs in the genome (> 20% minor allele frequency) have already been captured by these databases. The coverage drops dramatically below frequencies of 10%. At high frequencies, there is no sampling bias with respect to ethnicity or to regions of the genome. Finally, a relatively large fraction (> 40%) of SNPs in these databases were not seen in our study, which means that they are either of very low frequency, mismapped, or not polymorphic at all.