This research explores whether classroom life (CL), textbooks (TE), and learning initiative (LI) are mediators between instruction (P) and science performance, as well as whether these mechanisms are moderated by gender. 484 eighth grade students completed the questionnaire with four subscales of P, LI, TE, and CL. For the needs of triangulation and complementarity, three focus group interviews were conducted later. Based on mediation analysis and multi-group structural equation modeling, it was found that 1) the direct effects of P on LI, P on CL, P on TE, and LI on science performance are significant, while the other direct effects are insignificant; 2) comparing to the male group, the direct effect of P on LI in the female group is larger; 3) characteristics hindering students' science learning include: the pace of a lesson is too fast, pictures and experiments are less in the textbook, and top students and low proficiency students are uncooperative. Findings expose that instruction significantly influence students' science performance, and this impact is completely mediated by students' learning initiatives. The relation between instruction and learning initiative is stronger in the female group. Textbooks can be useless in the context that instruction does not match students' learning ability.