In Flanders, 2 classes per school were selected to participate in TIMSS. Therefore, a multilevel model with an intermediate level could be implemented. The data collection was extended with additional questionnaires and an intelligence test. The empty model revealed that 57% of the variance of mathematics scores was situated at the student level, 29% at the class level, and 14% at the school level. Furthermore, the results showed a significant reduction of the variance by intake characteristics, especially intelligence. The remaining student characteristics (attitude towards mathematics and the subject chosen) appear to have an additional effect.