Familial aggregation for disease is important; strong familial risk factors must exist even if the increased risk to a relative of an affected individual is modest. It is in practice difficult, however, to conduct studies in genetic epidemiology which conform to strict epidemiological principles. For twin studies there are two major questions: Are twins ‘no different’ from the population on which inference is to be made? Are study twins ‘no different’ to twins in the population? The importance of each question of bias depends on the scientific question, the trait(s) studied, and sampling issues. The strength of the twin design is its ability to refute the null hypothesis that genetic factors do not explain variation in a trait. Following the Popperian paradigm, alternate hypotheses should be considered in depth (both theoretically and empirically), with a design and sample size sufficient to exclude not just naive explanations. More sophisticated statistical techniques are now being applied, so the philosophy, assumptions, and limitations of statistical modelling must be appreciated. The concept of ‘heritability’ has, in the past, been misunderstood and misused. New advances in DNA technology promise to revolutionise epidemiological thinking, and so case-control-pedigree designs may well become standard tools. The strengths and limitations of studies based on related individuals as the sampling unit are discussed.