The measurement of behavior is inherently problematic, and this is especially true of substance use and abuse phenotypes. The contingent nature of many assessments, such that symptoms of abuse and dependence cannot be obtained from those who have not initiated substance use, presents special difficulties. Furthermore, it is not clear whether individual differences in liability to use, abuse and dependence are best characterized as dimensions or classes or a combination of both. This article outlines research designs and methods suitable for quantifying liability to substance use. The value of data collected from relatives is emphasized in this context, as they permit identification of models normally compromised because data on, e.g., substance dependence symptoms, are systematically missing in those who have not initiated use of the substance in question.