Using behaviorally discordant siblings to test for gene-behavior associations is a common tool in molecular genetics, because the within-family contrast offers a research design that avoids confounds inevitable in all between-family comparisons of unrelated individuals. We propose a similar strategy to assess the behavior-behavior associations on which much of psychological science is built. Between-family correlations of personality test scores (e.g., sensation seeking) and behavioral outcomes (e.g., substance use) may be mediated by variables that differ between families (e.g., social class or religiosity) and correlate with both personality and outcome. Contrasting twin and nontwin siblings who were highly discordant for behavioral correlates of substance use, we tested whether between-family behavioral correlations replicated within families. Some, but not all, did. Within-family analyses of behaviorally discordant siblings may find wide application in efforts to clarify the meaning of correlational research data.