Background. In several tracked educational systems, realizing optimal placements in classes in the first year of secondary education depends on the accuracy of teacher expectations. Aims. The aim of this study was to investigate between-teacher differences in their expectations regarding the academic aptitude of their students. Sample. The sample consisted of 500 teachers (classes) who provided their expectations of 7,550 students in the final grade of Dutch primary education. Methods. We analysed the extent to which teachers differed in their expectations and in what contexts their expectations were biased, using multilevel random slope models. Results. Multilevel analysis showed teacher expectation bias to be related to gender and socio-ethnic background of students. The differences among teachers in expectations for Turkish, Moroccan, and other foreign students with low-educated parents were larger than the average teacher expectation bias for these groups in the sample. When student characteristics were controlled for, we found that the teachers in our sample had higher expectations for students in high-performing classes or classes with only a small proportion of students from low-SES families. Teacher expectation bias for demographic groups, however, was found to be independent of the class population. Conclusions. The results indicate the importance of the teacher and the necessity of using multilevel models when investigating the complex nature of between-teacher differences in expectations of student performance.