Content and Language Integrated Learning (CLIL) is considered a promising approach to enhancing foreign language skills and motivation. However, its impact on students' academic self-concepts remains largely unclear. This study aimed to investigate whether CLIL positively affects students' English self-concepts but harms their math self-concepts in Grade 8 after two years of CLIL participation. Furthermore, the study intended to control for and disentangle selection and preparation effects caused by selective access and increased English instruction before the start of CLIL, as neglecting a priori differences between CLIL and non-CLIL students has led to overestimating CLIL effects in the past. Participants were 5963 academic-track school students. Propensity score matching was applied to control for selection effects. Structural equation modeling was used to estimate CLIL effects on English and math self-concepts. The inclusion of control variables allowed for accounting for preparation effects. CLIL students had significantly higher English self-concepts than non-CLIL students, which could be explained by selection and preparation effects. However, attending CLIL helped to maintain the advantage over non-CLIL students over the first two years of CLIL participation. CLIL had no detrimental effects on students' math self-concepts but left them unaffected. The study contributes to a deeper understanding of the effects of CLIL on students' self-concepts in different subjects. Furthermore, the results highlight the importance of accounting for both selection and preparation effects in future CLIL studies to obtain unbiased CLIL effect estimates. •First-time separation of selection, preparation, and CLIL effects on self-concepts.•Panel data, multiple covariates, and matching to control selection and preparation.•CLIL helps maintain, but not extend, pre-existing English self-concept advantages.•CLIL does not harm math self-concept.