Personalized normative feedback (PNF) interventions are generally effective at correcting normative misperceptions and reducing risky alcohol consumption among college students. However, research has yet to establish what level of reference group specificity is most efficacious in delivering PNF. This study compared the efficacy of a web-based PNF intervention using 8 increasingly specific reference groups against a Web-BASICS intervention and a repeated-assessment control in reducing risky drinking and associated consequences. Participants were 1,663 heavy-drinking Caucasian and Asian undergraduates at 2 universities. The referent for web-based PNF was either the typical same-campus student or a same-campus student at 1 (either gender, race, or Greek affiliation), or a combination of 2 (e.g., gender and race), or all 3 levels of specificity (i.e., gender, race, and Greek affiliation). Hypotheses were tested using quasi-Poisson generalized linear models fit by generalized estimating equations. The PNF intervention participants showed modest reductions in all 4 outcomes (average total drinks, peak drinking, drinking days, and drinking consequences) compared with control participants. No significant differences in drinking outcomes were found between the PNF group as a whole and the Web-BASICS group. Among the 8 PNF conditions, participants receiving typical student PNF demonstrated greater reductions in all 4 outcomes compared with those receiving PNF for more specific reference groups. Perceived drinking norms and discrepancies between individual behavior and actual norms mediated the efficacy of the intervention. Findings suggest a web-based PNF intervention using the typical student referent offers a parsimonious approach to reducing problematic alcohol use outcomes among college students.