Objective: Perceived descriptive drinking norms often differ from actual norms and are positively related to personal consumption. However, it is not clear how normative perceptions vary with specificity of the reference group. Are drinking norms more accurate and more closely related to drinking behavior as reference group specificity increases? Do these relationships vary as a function of participant demographics? The present study examined the relationship between perceived descriptive norms and drinking behavior by ethnicity (Asian or White), sex, and fraternity/sorority status. Method: Participants were 2,699 (58% female) White (75%) or Asian (25%) undergraduates from two universities who reported their own alcohol use and perceived descriptive norms for eight reference groups: "typical student"; same sex, ethnicity, or fraternity/sorority status; and all combinations of these three factors. Results: Participants generally reported the highest perceived norms for the most distal reference group (typical student), with perceptions becoming more accurate as individuals' similarity to the reference group increased. Despite increased accuracy, participants perceived that all reference groups drank more than was actually the case. Across specific subgroups (fraternity/sorority members and men) different patterns emerged. Fraternity/sorority members reliably reported higher estimates of drinking for reference groups that included fraternity/ sorority status, and, to a lesser extent, men reported higher estimates for reference groups that included men. Conclusions: The results suggest that interventions targeting normative misperceptions may need to provide feedback based on participant demography or group membership. Although reference group-specific feedback may be important for some subgroups, typical student feedback provides the largest normative discrepancy for the majority of students.