Previous studies on student well-being have focused on a limited number of factors. However, well-being is facilitated or hindered by many different factors. Therefore, focusing on a limited set of constructs could lead to an incomplete understanding of the various factors that predict student well-being. The current study drew on the Programme for International Student Assessment (PISA) dataset to understand the importance of background, non-cognitive/metacognitive, and schooling constructs in understanding well-being. This study focused specifically on understanding different well-being dimensions including positive affect, negative affect, life satisfaction, and eudaimonic well-being. The data were from 12,058 15-year-old Chinese students from Beijing, Shanghai, Jiangsu, and Zhejiang. China presents an interesting case given its high levels of achievement but low levels of well-being. Using a machine learning approach (i.e., random forest regression), the results indicated that factors belonging to “non-cognitive/metacognitive” and “schooling” constructs were found to be the most important predictors of well-being. More specifically, students’ positive affect and life satisfaction were best predicted by school belonging and resilience. Negative affect was best accounted for by school belonging and fear of failure. Eudaimonic well-being was best predicted by resilience and work mastery. Theoretical and practical implications are discussed.