This issue of the International Journal of Epidemiology reprints a seminal letter to the editor by Martijn Katan,1 which appears to be the first description of the concept of ‘Mendelian randomization.’ In discussing the controversy over whether the association between low serum cholesterol and cancer is causal or might simply reflect an effect of the disease to lower cholesterol levels (‘reverse causation’) or confounding by diet or other factors, Katan proposed a test of causality by studying instead the relationship between cancer and a genetic determinant of serum cholesterol, the apolipoprotein A (APOE) gene. His rationale was that since alleles are allocated essentially at random, such an association would not be subject to either confounding or reverse causation. Thus, if a causal relationship between APOE and serum cholesterol were clearly established, then an association between APOE and cancer would provide indirect evidence for the causality of the association between serum cholesterol and cancer. Although Katan did not use the term ‘Mendelian randomization’, the concept has been attributed to him and subsequently developed by a number of other authors.2–6 In particular, Davey Smith and Ebrahim2 have shown how the magnitude of the estimated effects of a gene (G) on an intermediate phenotype (IP) and on disease (D) can be combined to yield an estimate of the causal effect of the intermediate phenotype on disease, as illustrated in the following figure: (where the dotted arrow from G to D represents the indirect association assumed to be mediated entirely through IP).