I brought together two strands of literature, one investigating the moderate but persistent underachievement of girls in mathematics in most countries, and the other examining the role of self-efficacy and other self-beliefs in predicting behaviour and achievement. I implemented detailed decompositions of the gender mathematics gap, both at the mean and for low and high performing students, for a large and diverse group of countries. I found considerable heterogeneity and different cross-country patterns in decomposition components and the contribution of self-beliefs. In OECD-Europe and more affluent East Asian countries, most or all of the gap is explained by gender differences in self-beliefs, especially self-efficacy; on the other hand, in Latin America and the Middle East, most of the gap remains unexplained. I also investigated the cross-country relationship between the gender mathematics gap and gender equity and found that a clearly negative association can be established after controlling for cross-country heterogeneity in gender differences in mathematics self-beliefs, which correlate with gender equality.