Abstract The design of learning environment studies investigating students’ perceptions often is multilevel in nature. This multilevel nature of studies can appear in the object of research (for example, teacher behaviour towards the individual student or towards the class), the level of perception (personalised perceptions or group perceptions) and the sampling of data (usually clustered: students are sampled with their classmates, classes are sampled with other classes taught by the same teacher, etc.). In the present study, the impact of decisions about level is studied using students’ perceptions of the teacher–student relationship as assessed with the Questionnaire on Teacher Interaction (QTI). Data were gathered in one school (59 classes of 29 teachers) with two versions of this questionnaire: a personalised version and a class version. For reasons of comparison, additional data on the class version were analysed from 44,415 students from 1,913 teachers in 207 schools. Results from multilevel and single-level analyses of the class and personal versions of the QTI indicated that multilevel analyses are to be preferred over single-level analyses and that different conceptual structures could apply depending on the object of study and the level of perception.