When using school performance indicators for the purposes of accountability and school choice, it is very important that measures are stable over time. This issue is very relevant when the system is based on the allocation of students to the same teachers for the whole educational cycle (e.g. at primary school level). Using administrative data provided by the National Evaluation Committee for Education (INVALSI), we explore the stability of performance estimates for Italian primary schools. We first construct school performance metrics using INVALSI standardised tests, by taking advantage of a rich array of individual‐level variables (including prior achievement) that allow us to estimate a school effect in a ‘value added’ perspective. We assess differences in performance across schools and the persistence of these differences across cohorts. We find that controlling for compositional differences is important; therefore, estimates of school performances are unstable across metrics, and correlations across cohorts decline over time.