This article presents the findings from a teacher intervention in Dutch kindergartens aimed at improving teachers' differentiation practices (DP) to better anticipate student differences. The intervention was designed to improve the match between student levels and curricular activities, in particular for high-ability students and consists of three components. A pretest-posttest cluster randomised design was used with three conditions: control (n = 34), pilot intervention (n = 32) and improved intervention (n = 34). Quantitative results demonstrate that the intervention fidelity was relatively high in the improved intervention. Correlations confirmed that teacher-reported DP were higher in schools where the intervention was more completely implemented. DP were enhanced in both intervention conditions, but showed greater improvement in the improved intervention. Qualitative data revealed process characteristics that reflect problems schools encounter with this intervention. Teachers can be supported in improving DP, but this requires school-wide intensive and long support.