Confounding is a basic problem of comparability--and therefore has always been present in science. Originally a plain English word, it acquired more specific meanings in epidemiologic thinking about experimental and non-experimental research. The use of the word can be traced to Fisher. The concept was developed more fully in social science research, among others by Kish. Landmark developments in epidemiology in the second half of the 20th century were by Cornfield and by Miettinen. These developments emphasised that reasoning about confounding is almost entirely an a priori process that we have to impose upon the data and the data-analysis to arrive at a meaningful interpretation. The problems of confounding present their old challenges again in recent applications to genetic epidemiology.