I explored the effect of socio-economic status (SES) on mathematics achievement by gender at grades 5 in 2012 and 10 in 2017. Secondary Italian data (30,868 pupils in primary and 38,091 in secondary school) were analysed via the Rasch model to estimate students' ability in mathematics and then via a three-level regression (students nested into classroom and schools) to estimate SES effect (calculated at individual, classroom and school level) on ability. Data analysis showed positive correlation between SES and ability, and a significant mediating role of gender, in secondary more strongly than in primary school, and a greater effect of classroom and school rather than individual SES on students attending secondary school. In particular, at grade 10, individual SES effect tends to disappear when classroom and school SES are accounted for, suggesting that adolescents are much more prone to the effect of context variables than younger pupils are.