Objective: College students are among the age-group most affected by sexually transmitted infections, yet few evidence-based interventions have proved effective in increasing college student condom use. The goal of this study was to determine which combination of referent proximity and width of discrepancy between perceived and actual norms within a normative feedback intervention produced the greatest motivation to increase condom use among US college students. Design: A 2 (proximity) × 2 (width of discrepancy) × 2 (gender) randomised-factorial experiment was conducted. Setting: Sexually active college students (N = 212; 50.5% female; 70.4% White) were recruited via Amazon Mechanical Turk, an online labour market. Method: Participants were randomised, stratified by gender, to one of four sham feedback conditions--proximal referent wide discrepancy, proximal referent narrow discrepancy, distal referent wide discrepancy and distal referent narrow discrepancy. Following delivery of the feedback, participants completed measures assessing their willingness to change their condom use. Results: A three-way factorial analysis of covariance revealed a significant interaction between referent proximity and width of discrepancy (F = 7.88, p = 0.005, [partial eta squared] = 0.04), such that the effect of proximity on willingness to use condoms was greater in the narrow, compared with wide, discrepancy condition. Conclusion: Findings from this study indicate that it may be beneficial to assess students' perceptions of their peers' sexual behaviour before selecting the reference group to include within normative feedback. This is the first study to demonstrate that the importance of selecting referents within normative feedback may be dependent on the accuracy of students' perceptions.