Objective: We studied how representative cytologically abnormal women (“referral populations”) are with respect to uncovering differences between human papillomavirus (HPV) assays in the primary screening where most women are cytologically normal. Methods: A total of 4997 women were tested with SurePath® cytology, and Hybrid Capture 2 (HC2), cobas, CLART and APTIMA HPV assays. Women with positive test results were offered a follow-up. For all detected HPV infections and HPV-positive high-grade cervical intraepithelial neoplasia (≥CIN2), we studied the distributions of assay-specific signal strengths in the baseline samples as documented by the assays’ automatically generated reports. We calculated the likelihood of test result discordance as the proportion of HPV-positive samples that were not confirmed by all four assays. Results: Median signal strengths for HPV infections were weaker in normal than abnormal cytology (P<.001, adjusted for women's age, multiple infections and the reason for taking the sample). For HC2, they were RLU/CO 11.0 (interquartile range, IQR: 3.3-52.8) vs 124.2 (IQR: 22.8-506.9), respectively; for cobas, Ct 33.5 (IQR: 29.6-37.5) vs 26.9 (IQR: 23.7-31.3), respectively; for APTIMA, S/CO 10.2 (IQR: 5.8-11.3) vs 11.1 (IQR: 9.4-15.5), respectively. Similar patterns were observed for HPV-positive ≥CIN2. The four HPV assays more frequently returned discordant test results in normal than in abnormal cytology. Relative frequency of discordance in detecting HPV infections was 0.39 (95% confidence interval: 0.33-0.48) for abnormal vs normal cytology. Conclusions: These data suggest that referral population studies, by not including sufficient numbers of cytology normal women, underestimate the differences between HPV assays that would become apparent in primary screening.