Background: The differential effectiveness of schools and teachers receives a growing interest, but few studies focused on the relevance of student ethnicity for this effectiveness and only a small number of these studies investigated teaching in terms of the teacher-student interpersonal relationship. Furthermore, the methodology employed often restricted researchers to investigating direct effects between variables across large samples of students. Aims: This study uses causal modelling to investigate associations between student background characteristics, students' perceptions of the teacher-student interpersonal relationship, and student outcomes, across and within several population subgroups in Dutch secondary multi-ethnic classes. Methods and sample: Multi-group structural equation modelling was used to investigate causal paths between variables in four ethnic groups: Dutch (N = 387), Turkish first- and second-generation immigrant students (N = 267), Moroccan first and second generation (N = 364), and Surinamese second-generation students (N = 101). Results: Different structural paths were necessary to explain associations between variables in the different (sub) groups. Different amounts of variance in student attitudes could be explained by these variables. Conclusions: The teacher-student interpersonal relationship is more important for students with a non-Dutch background than for students with a Dutch background. Results suggest that the teacher-student relationship is more important for second generation than for first-generation immigrant students. Multi-group causal model analyses can provide a better, more differentiated picture of the associations between student background variables, teacher behaviour, and student outcomes than do more traditional types of analyses.