The purpose of this study was to investigate the degree to which A. Bandura's (1997) hypothesized sources of self-efficacy predict the science self-efficacy beliefs of middle school students (N = 319), to replicate previous findings that science self-efficacy predicts science achievement, and to explore how science self-efficacy and its antecedents differ by gender. Significant correlations were found between mastery experiences, vicarious experiences, social persuasions, physiological arousal, and self-efficacy. Only mastery experiences significantly predicted science self-efficacy. Girls reported stronger science self-efficacy than did boys. Findings support and extend the theoretical tenets of Bandura's social cognitive theory. © 2006 Wiley Periodicals, Inc. J Res Sci Teach 43: 485–499, 2006