ABSTRACT The review presented here discusses and exemplifies problems in epidemiological studies of drug teratogenesis according to methodology: case–control studies, cohort studies, or total population studies. Sources of errors and the possibility of confounding are underlined. The review stresses the caution with which conclusions have to be drawn when exposure data are retrospective or other possible bias exists. It also stresses the problem with the multiple testing situation that is usually present in the studies. It is therefore difficult to draw any firm conclusion from single studies and still more difficult to draw conclusions on causality. As randomized studies are in most cases out of the question, one has to rely on the type of studies which can be made, but the interpretation of the results should be cautious. The ideal study, next to a randomized one, is a large prospective study with detailed exposure information and detailed and unbiased outcome data. Even so, such a study can mainly be used for identifying possible associations which have to be verified or rejected in new studies. Nearly every finding of a risk increase, if not extremely strong, should only be regarded as a tentative signal to be tested in independent studies.