In the Advanced Online Publication of Nature Genetics, John Ioannidis and colleagues at the University of Ioannina School of Medicine, Ioannina, Greece, describe a study to determine the reliability of disease association and genetic linkage reports (DOI:10.1038/ng749). They assembled data from published reports of 36 different disease associations, ranging from schizophrenia to hypertension. They used meta-analysis to explore the diversity and discrepancies between different studies. In 39% of cases they found statistically relevant heterogeneity between studies. The first published reports (often appearing in high impact-factor journals) tended to give more impressive disease association, which became less prominent in subsequent reports. In some cases the accumulation of additional data led to increased statistical significance of the genetic association. Ioannidis et al. conclude that meta-analysis is useful for the cautious evaluation of genetic association studies and human genome epidemiology.