Student ratings of teaching quality have been shown to be powerful predictors of important academic outcome variables. This is the case despite the fact that students from the same classroom can perceive teaching quality quite differently in their own idiosyncratic ways. These differences among students in the same classroom are typically dismissed as the result of individual rating tendencies (e.g., rater leniency) with only low relevance for students’ learning. However, such idiosyncratic perceptions might also be the result of meaningful differences among students in that they could reflect dyadic effects between individual students and specific teachers (e.g., a trusting relationship). In the present study, we examined students’ perceptions of teaching quality in mathematics lessons (structure, monitoring, comprehensibility, and support) in a nationally representative sample of 204 German secondary school mathematics classes. We used a quasi-experimental design for students who had the same teacher in Grades 9 and 10 and students who had a new teacher in Grade 10 to differentiate between students’ time-consistent rating tendencies and dyadic student-teacher effects. The results of latent variable models showed that student rating tendencies were found for all quality dimensions, with the highest impact observed for students’ general impression and teacher monitoring. The results for students with the same teacher showed additional dyadic effects. Students’ general impression and comprehensibility exhibited greater time consistency when students rated the same teacher than when students rated different teachers. Finally, students’ rating tendencies and dyadic effects were related to students’ outcomes.