Though schooling is robustly associated with children’s cognitive performance gains, the magnitude of its effect remains unclear because access to quality schooling covaries with socioeconomic status (SES), health, and other environmental factors. Utilizing a natural experiment, we assess children’s abstract reasoning, widely considered predictive of academic achievement, longitudinally in a rural Bolivian subsistence population with considerable variability in school quality, but minimal variability in SES and child nutritional status. Among Tsimane forager-horticulturalists (n=290 children aged 8-18 years; 44% female), we use Raven’s Coloured Progressive Matrices (RCPM) to evaluate whether abstract reasoning is predicted by school quality and individual characteristics including attendance, school-derived skills (i.e. reading ability, arithmetic ability), child nutritional status (i.e. weight-for-height), and family-level factors including parental education and wealth. We also evaluate predictors of prospective change in children’s RCPM score after a brief training session approximately one month following baseline RCPM assessment, and four years later. Relative to children attending low quality schools, children attending higher quality schools (e.g. government trained teachers, classroom materials) have greater reading ability, score higher on RCPM at baseline, and show greater improvement in RCPM score after training. Reading ability and school attendance positively predict baseline RCPM score and their relationship to RCPM is moderated by school quality. School quality and weight-for-height have the largest effect prospectively on RCPM score among children who stayed in school throughout the study. These findings suggest school quality is a more important causal determinant of abstract reasoning than school attendance or certain scholastic skills.