Purpose: In school improvement studies, randomized experiments are rare. A special problem is the assignment to the experimental and control groups, taking into account the different starting conditions at the schools in terms of school improvement competencies. The purpose of this paper is to take the example of an intervention study conducted in Baden-Württemberg, Germany, the extent to which the challenges involved with a quasi-experimental design were addressed is examined. Design/methodology/approach: The intervention study was conducted with 54 math teachers (experimental group: n=29; control group: n=25) and their grade 7 and 8 classes (n=1,054) at 13 secondary schools. It aimed to increase teacher cooperation on teaching for promotion of students' self-regulated learning. T-tests, Mann-Whitney tests, ANOVA, multilevel regression analyses were conducted. Findings: At the beginning of the intervention, the teachers in the two groups did not differ significantly in prior cooperation on teaching processes and in attitudes toward cooperation. However, they differed in prior cooperation on school framework conditions and teaching processes. The intervention was effective in increasing teachers' cooperation intensity on instruction, and teachers' attitude toward binding cooperation. However, teaching processes did not change depending on experimental or control group. Research limitations/implications: Teacher cooperation practice was assessed only by teachers' self-report. No indicators on the quality of the cooperation among teachers were included. Practical implications: The paper discusses the challenges and limitations of conducting intervention studies on school improvement. Implications for further research are given. Originality/value: This paper fulfils an identified need to study how quasi-experimental designs can be implemented in intervention studies on school improvement.