Background: In many countries and regions, such as the United States, Europe and China, a trend has emerged in which students' enthusiasm for STEM is declining. This decline may be related to students' lack of science self-efficacy. An accurate examination of students' science self-efficacy can provide a research foundation for how to cultivate it. This paper used mixed methods to develop a valid science self-efficacy scale for high school students, focusing on the perceived competence dimension. A cross-sectional analysis exploring and interpreting differences across grades and genders in science self-efficacy among high school students was conducted. Subsequently, a 1-year longitudinal study was conducted on the development of science self-efficacy in China. Results: This study developed a 24-item science self-efficacy instrument based on the Rasch model, and the validity of the instrument was assessed through multiple aspects, including face, content, construct, and predictive validity. This instrument was used to divide students' science self-efficacy into four different levels. A cross-sectional study examining 1564 high school students in 10th--12th grades revealed that students' science self-efficacy exhibited a complex process of decreasing and then increasing by grade. Most girls' science self-efficacy was higher than that of boys for Levels 1 and 4, while for the intermediate levels, i.e., Levels 2 and 3, most boys had higher science self-efficacy than girls. The quantitative and qualitative results of the longitudinal study through a 1-year follow-up of 233 high school students indicated that students' science self-efficacy significantly improved. We revealed inconsistencies between cross-sectional and longitudinal studies of the change in science self-efficacy from 10 to 11th grade. Conclusion: This study makes many contributions. First, we developed a science self-efficacy measurement instrument for high school students with high reliability and validity based on the Rasch model and characterized four different levels of student science self-efficacy. Second, the gender differences in science self-efficacy and the complex changes among grades were explained from the perspective of science self-efficacy level. Finally, students' science self-efficacy significantly improved in the longitudinal study, which was explained by self-efficacy theory and the Chinese core competency-oriented science curriculum.