Based on Bandura's social cognitive theory, this study was designed to examine positive and negative outcome expectancy and refusal self-efficacy of Internet use and their contribution to Internet addiction among college students by using hierarchical multiple regression analyses in a cross-sectional study design. Schools were first stratified into technical or nontechnical colleges and then into seven majors. A cluster random sampling by department was further applied to randomly choose participants from each major. A representative sample of 4,456 college students participated in this study. The Outcome Expectancy and Refusal Self-Efficacy of Internet Use Questionnaire and the Chen Internet Addiction Scale were used to assess the cognitive factors and the levels of Internet addiction. Results showed that both positive outcome expectancy and negative outcome expectancy were significantly and positively correlated with Internet addiction, and refusal self-efficacy of Internet use was significantly and negatively related to Internet addiction. Further analyses revealed that refusal self-efficacy of Internet use directly and negatively predicted Internet addiction. Moreover, we discovered that positive outcome expectancy positively predicted Internet addiction via refusal self-efficacy of Internet use; however, surprisingly, negative outcome expectancy had both a direct and indirect positive relationship in predicting Internet addiction via the refusal self-efficacy of Internet use. These results give empirical evidence to verify the theoretical effectiveness of the three cognitive factors to Internet addiction and should be incorporated when designing prevention programs and strategies for Internet addicted college students.