We examined the predictability of non-cognitive variables for students' mathematics achievement, based on large-scale international databases of the TIMSS 2003, 2007, and 2011, and the PISA 2003 and 2012. We synthesized empirical evidence about 65 non-cognitive variables, which were categorized into 13 research domains of educational psychology—affect, curriculum/content exposure, homework, learning and instructional time, motivation, personality traits, planned behavior, school climate, self-beliefs/social-cognitive theory, self-regulatory learning style/strategies, teacher behavior, value, and vocational interest. Our analyses showed that a group of self-beliefs constructs, in particular, self-efficacy in PISA, confidence in TIMSS, and educational aspiration, in both TIMSS and PISA, were the best predictors of individual-level student achievement in mathematics. The present review supports the claim that students' projective judgements about their own ability and future selves are particularly important for their academic achievement. We discuss potential educational initiatives to maximize educational outcomes of students from diverse cultural and national backgrounds. •Self-beliefs are the best predictors of individual-level student achievement.•The effect of self-efficacy belief is stronger than that of SES at the individual-level.•SES was the strongest, positive predictor of country-level student achievement.