Traditional science, technology, engineering, and mathematics (STEM) education is sometimes criticized for lacking approaches to present real-world practices and phenomena beyond naked eyes. Head-mounted display virtual reality (HMD VR) provides opportunities to solve this issue. However, little is known about the impact of this approach on student's self-efficacy in science. This study is to address this knowledge gap. Sixty-six 11th grade students were recruited to participate in an HMD VR learning activity. Half of these students filled in a science self-efficacy questionnaire before the VR learning activity, and the others filled in it after the activity. The study compared (1) students' science self-efficacy between these two conditions and (2) students' post-activity science self-efficacy among different science-class grading score groups. Results showed that the change of students' science self-efficacy was not significant after the learning activity and the differences among most science-class grading score groups were small. After the results were analyzed, the capability of affording gestures and physical movement was recognized as an important factor that determined whether an HMD VR learning environment could significantly enhance students' science self-efficacy; educators were suggested to not use science class scores to predict students' potential and future achievements in science.