Our analyses of Carey's (1992) simulated data set of substance abuse in a cohort of adolescent twins were aimed at answering the question What is the relationship between age at first drug use and EVER having used drugs (i.e., teenage drug use liability)? Three analytic methods were used to determine whether age at first drug use was (1) a "perfect" index of drug use liability, (2) correlated in relatives but conditionally independent of drug use liability, or (3) causally influenced by drug use liability and by factors independent of liability. The analytic methods included nonmetric multidimensional scaling, multifactorial threshold model-fitting to contingency tables, and pedigree-based likelihood formulations for the raw data. All approaches indicated that age at first drug use was a perfect index of drug use liability. Further, model-fitting results indicated that only shared environmental factors accounted for twin similarity in the onset and timing of drug use. We discuss the limitations of each of the analytic methods and integrate our findings with the true model used in Carey's simulation.