Identifying meaningful cognitive and noncognitive predictors of mathematical competence is critical for developing targeted interventions for students struggling with mathematics. Here, 317 students' short-term verbal memory, verbal and visuospatial working memory, complexspatial abilities, intelligence, and mathematics attitudes and anxiety were assessed, and their teachers reported on their attentive behavior in 7th-grade mathematics classrooms. Bayesian regression models revealed that complex spatial abilities and in-class attention were the most plausible predictors of 7th-grade mathematics, but not word reading achievement, controlling for prior mathematics achievement. These results were confirmed with multilevel models that revealed interactions between these factors and prior achievement. The largest gains were among students with strong mathematical competencies in 6th grade, and average or better in-class attention in 7th grade as well as above average spatial abilities. High mathematics anxiety was associated with lower attention and through this indirectly influenced achievement gains. These results have implications for how to best target interventions for students at risk for long-term difficulties with mathematics.