We look at the effect of school starting age on standardized test scores using data covering all grade four and grade eight students in Hungary. Instrumental variables estimates of the local average treatment effect suggest that children generally gain from starting school one year later and the effects are much stronger in the case of students coming from low-educated families. We test the robustness of the results by allowing for heterogeneity in the age effect, distinguishing between fields of testing, using discontinuity samples and relying on alternative data. The hypothesis that delayed entry has a stronger impact on low-status children is supported by the robustness checks. The observed patterns are most probably explained by the better performance of kindergartens, as opposed to schools, in developing the skills of low-status children.