Background: Students are vulnerable to Internet addiction (IA). Influences of cognitions based on the Health Belief Model (HBM) and perceived number of peers with IA (PNPIA) affecting students' IA, and mediating effects involved, have not been investigated. Methods: This cross-sectional study surveyed 9518 Hong Kong Chinese secondary school students in the school setting. Results: In this self-reported study, the majority (82.6 %) reported that they had peers with IA. Based on the Chinese Internet Addiction Scale (cut-off =63/64), the prevalence of IA was 16.0 % (males: 17.6 %; females: 14.0 %). Among the non-IA cases, 7.6 % (males: 8.7 %; females: 6.3 %) perceived a chance of developing IA in the next 12 months. Concurring with the HBM, adjusted logistic analysis showed that the Perceived Social Benefits of Internet Use Scale (males: Adjusted odds ratio (ORa) = 1.19; females: ORa = 1.23), Perceived Barriers for Reducing Internet Use Scale (males: ORa = 1.26; females: ORa = 1.36), and Perceived Self-efficacy for Reducing Internet Use Scale (males: ORa = 0.66; females: ORa = 0.56) were significantly associated with IA. Similarly, PNPIA was significantly associated with IA ('quite a number': Males: ORa = 2.85; females: ORa = 4.35; 'a large number': Males: ORa = 3.90; females: ORa = 9.09). Controlling for these three constructs, PNPIA remained significant but the strength of association diminished ('quite a number': Males: Multivariate odds ratio (ORm) = 2.07; females: ORm = 2.44; 'a large number': Males: ORm = 2.39; females: ORm = 3.56). Hence, the association between PNPIA and IA was partially mediated (explained) by the three HBM constructs. Interventions preventing IA should change these constructs. Conclusions: In sum, prevalence of IA was relatively high and was associated with some HBM constructs and PNPIA, and PNPIA also partially mediated associations between HBM constructs and IA. Huge challenges are expected, as social relationships and an imbalance of cost-benefit for reducing Internet use are involved. Perceived susceptibility and perceived severity of IA were relatively low and the direction of their associations with IA did not concur with the HBM. Group cognitive-behavioral interventions involving peers with IA or peers recovered from IA are potentially useful to modify the HBM constructs and should be tested for efficacy.