This paper examines which conguration of teaching activities (expressed in, e.g., problem solving, homework, lecturing) maximizes student performance. To do so, it formulates a non-parametric eciency model that is rooted in the Data Envelopment Analysis literature. In the model, we account for (1) self selection of students and teachers in better schools, and (2) complementary teaching activities. The analysis distinguishes both individual teaching (i.e., a personal teaching style adapted to the individual needs of the student) and collective teaching (i.e., a similar style for all students in a class). Exploiting the data set, we compare the actual teaching style as revealed by the teacher in the data to the model estimations. As such, we anal-yse which students in the class the teacher is targeting with his/her teaching style. The main results show that high test scores are associated with teaching styles that emphasise problem solving and homework. In addition, teachers seem to adapt their optimal teaching style on the 70 percent least performing students.