The efficient design of association mapping studies relies on a knowledge of the rate of decay of linkage disequilibrium with distance. This rate depends on the population recombination rate, C. An estimate of C for humans is usually obtained from a comparison of physical and genetic maps, assuming an effective population size of approximately 104. We demonstrate that under both a constant population size model and a model of long-term exponential growth, there is evidence for more recombination in polymorphism data than is expected from this estimate. An important contribution of gene conversion to meiotic recombination helps to explain our observation, but does not appear to be sufficient. The occurrence of multiple hits at CpG sites and the presence of population structure are not explanations.