In this article, different inspection models are compared in terms of their impact on school improvement and the mechanisms each of these models generates to have such an impact. Our theoretical framework was drawn from the programme theories of six countries’ school inspection systems (i.e. the Netherlands, England, Sweden, Ireland, the province of Styria in Austria and the Czech Republic). We describe how inspection models differ in the scheduling and frequency of visits (using a differentiated or cyclical approach), the evaluation of process and/or output standards, and the consequences of visits, and how these models lead to school improvement through the setting of expectations, the use of performance feedback and actions of the school’s stakeholders. These assumptions were tested by means of a survey of principals in primary and secondary schools in these countries (n = 2239). The data analysis followed a three-step approach: (1) confirmatory factor analyses, (2) path modelling and (3) fitting of multiple-indicator multiple-cause models. The results indicate that Inspectorates of Education that use a differentiated model (in addition to regular visits), in which they evaluate both educational practices and outcomes of schools and publicly report inspection findings of individual schools, are the most effective. These changes seem to be mediated by improvements in the schools’ self-evaluations and the schools’ stakeholders’ awareness of the findings in the public inspection reports. However, differentiated inspections also lead to unintended consequences as principals report on narrowing the curriculum and on discouraging teachers from experimenting with new teaching methods.