Background: Adaptive computerized interventions may help improve preterm children’s academic success, but randomized trials are rare. We tested whether a math training (XtraMath®) versus an active control condition (Cogmed®; working memory) improved school performance. Training feasibility was also evaluated. Methods: Preterm born first graders, N = 65 (28–35 + 6 weeks gestation) were recruited into a prospective randomized controlled multicenter trial and received one of two computerized trainings at home for 5 weeks. Teachers rated academic performance in math, reading/writing, and attention compared to classmates before (baseline), directly after (post), and 12 months after the intervention (follow-up). Total academic performance growth was calculated as change from baseline (hierarchically ordered—post test first, follow-up second). Results: Bootstrapped linear regressions showed that academic growth to post test was significantly higher in the math intervention group (B = 0.25 [95% confidence interval: 0.04–0.50], p = 0.039), but this difference was not sustained at the 12-month follow-up (B = 0.00 [−0.31 to 0.34], p = 0.996). Parents in the XtraMath group reported higher acceptance compared with the Cogmed group (mean difference: −0.49, [−0.90 to −0.08], p = 0.037). Conclusions: Our findings do not show a sustained difference in efficacy between both trainings. Studies of math intervention effectiveness for preterm school-aged children are warranted. Impact: Adaptive computerized math training may help improve preterm children’s short-term school performance.Computerized math training provides a novel avenue towards intervention after preterm birth.Well-powered randomized controlled studies of math intervention effectiveness for preterm school-aged children are warranted.