This paper estimates peer age effect on educational outcomes of Italian pupils attending primary school by exploiting changes in enrollment rules over the last few years. The empirical procedure allows to understand if there is selection in classroom formation, arguing that in the absence of pupils sorting by early age at school entry, it is possible to estimate the "true" peer age effect. Results suggest that the proportion of youngest students "in advance" in the classroom has a positive impact on child's achievements measured by Normalized and Rasch test scores both in Reading and Mathematics. Additional empirical evidence shows that the effect on individual scores of sharing the classroom with pupils "in advance" differs by students' age group and it is higher for youngest pupils. Findings do not seem to be invalidated by "cheating" phenomena that may occur during the implementation and evaluation of tests.