The education system strives to help students from low-income families achieve academic success. Academic resilience is related to not only individuals but also classrooms and schools. This study aimed to construct a comprehensive resilience model in science domains that presents the image of resilient students and describes the mechanisms by which they arise. This goal was achieved through a parametric multilevel latent profile analysis with level 1 and level 2 covariates. Latent profile indicators included variables reflecting motivation, subjective well-being, and engagement in science of secondary school students. Variables related to science classroom and school contexts constituted the levels 1 and 2 covariates. We used secondary data from Japan and Macao (China) collected in the 2015 Programme of International Student Assessment (PISA). The sample sizes were 1594 and 1114, respectively. We identified three resilience levels among students. The emergence of resilient students benefited from science classroom discipline, teacher-directed instruction, inquiry-based teaching, adjusting instruction, and teacher support. In Japan, the most robust predictor was inquiry-based teaching, while in Macao, it was teacher-directed instruction. Therefore, science teachers play a crucial role in building students' resilience. It also suggests that a small class decreases the likelihood of students entering the high-level resilience group.