Familial resemblance, which arises when members within families are more similar than are unrelated pairs of individuals, may be estimated in terms of correlations (or covariances) among family members. The magnitudes of such correlations generally reflect both the extent of environmental sharing and the degree of biological relationship between the relatives. Heritability, or more appropriately multifactorial heritability or generalized heritability, quantifies the strength of the familial resemblance and represents the percentage of variance in a trait that isdue to all additive familial effects including additive genetic effects and those of the familial environment. However, the traditional concept of heritability, which may be more appropriately called the genetic heritability, represents only the percentage of phenotypic variance due to additive genetic effects. Resolving the sources of familial resemblance entails other issues. For example, there may be major gene effects that may be largely or entirely nonadditive, temporal or developmental trends, and gene-gene (epistasis) and gene-environment interactions. The design of a family study determines which of these sources are resolvable. For example, in intact nuclear families consisting of parents and offspring, the genetic and familial environmental effects are not resolvable because these relatives share both genes and environmental effects are not resolvable because these relatives share both genes and environments. However, extended pedigrees and twin and adoption study designs allow separation of the heritable effects and, possibly, more complex etiologies, including interactions. Various factors affect the estimation and interpretability of heritability, for example, assumptions regarding linearity and additivity of the effects, assortative mating, and the underlying distribution of the data. Nonnormality of the data can lead to errors in hypothesis testing, although it yields reasonably unbiased estimates. Fortunately, these and other complications can be directly modeled in many of the sophisticated software packages available today in genetic epidemiology.