A common procedure for assessing children's behavior is to obtain parental ratings of the child. Since the ratings obtained are a function of both parent and child, disentangling the child's phenotype from that of the rater becomes an important methodological problem. For the analysis of genetic and environmental contributions to children's behavior, solutions to this are available when multiple raters, e.g., two parents, rate multiple children, e.g., twins. This paper describes and illustrates simple LISREL models for the analysis of parental ratings of children's behavior. We show how the assumption that mothers and fathers are rating the same behavior in children can be contrasted with the weaker alternative that parents are rating correlated behaviors. Given the stronger assumption, which appears adequate for ratings of children's internalizing behavior problems, the contribution of rater bias and unreliability may be separated from the shared and nonshared environmental components of variation in a behavior genetic analysis.