South Africa participated in the Third International Mathematics and Science Study (TIMSS) in 1995 and again in 1999 (TIMSS-Repeat) and in both studies the performance was extremely low compared to the other countries in the studies. In both studies more than 70% of the pupils wrote the achievement tests in their second or third language. A national option, an English test, was included together with the TIMSS-R mathematics and science tests in an attempt to ascertain the level of the pupils' language proficiency. Furthermore, additional questions pertaining to the pupils and their teachers' exposure and usage of English both within and outside of school were also included in the background questionnaire. In this research project that included more than 8 000 pupils in 200 schools, all the items pertaining to English proficiency and language usage and their relationship to mathematics achievement were explored. Partial Least Square analysis was used to explore the relative contribution of these factors to pupils' achievement together with other background variables from the student, teacher and principal questionnaires, resulting in the presentation of a schoollevel, a classroom-level model, a student-level model and a combined class and school-level model. Multi-level analysis was employed whereby a 2-level model (school and class-level and the student-level) was analysed in order to investigate the main factors explaining achievement of South African pupils in mathematics. The study revealed that the pupils' proficiency of English was a strong predictor of their success in mathematics. A number of other background variables on student and class-level were found to be significant. However, home language and class size were amongst those that were not found to have significant effect on achievement, whilst the effect of socio-economic status had a lesser effect once certain class-level factors were taken into consideration.