Different approaches for measuring instructional quality have been debated. Previous studies based on student ratings have primarily used class-average ratings. Beyond this, the high within-classroom variability of students' ratings might indicate that instruction caters to some, but not all students. Therefore, we investigated student-reported instructional quality in mathematics classrooms by considering the average student ratings and rating heterogeneity within classrooms. Using a case-centered clustering approach, we aimed to detect meaningful configurations of the level and heterogeneity of student-reported instructional quality in terms of the Three Basic Dimensions (TBD): classroom management, cognitive activation, and student support. We analyzed data from N = 973 grade eight students across N = 106 classes. Using Latent Profile Analysis (LPA), we identified four classroom profiles comprising 20% to 28% of the sample. The results indicate that the profile with the lowest average ratings showed consistently high heterogeneity for all indicator variables. However, the profile with the highest average ratings exhibited consistently low heterogeneity. We gained interesting insights into between-class differences in instructional quality by considering rating heterogeneity. Furthermore, we explored how classrooms from the identified profiles differed regarding socio-economic status (SES) and mathematics-related characteristics (interest, intrinsic motivation, self-concept, and achievement).