This study evaluates the effects of specialized educational programs for linguistic minorities, focusing on bilingual indigenous schools in Mexico, on academic performance in math and Spanish and on bullying. These schools aim to provide instruction in native languages and create safe spaces for minorities who often face discrimination; however, they encounter significant implementation challenges. I develop and estimate a structural model of parents choosing primary schools for their children that incorporates bullying as a key social interaction depending on the school ethnic composition and heterogeneous human capital formation technologies. I find that bullying consistently negatively impacts academic performance (by 0.15-0.38 standard deviations) and that increasing the proportion of indigenous students in a school from 0% to 100% reduces bullying for indigenous students by 7.3 percentage points, without affecting non-indigenous students. However, I find that teachers in indigenous schools are on average less effective than teachers in regular schools in promoting academic achievement. Their effectiveness is mediated by their indigenous language proficiency. Parental school choice is influenced both by academic aspirations and bullying concerns. I use the estimated model to evaluate counterfactual policies. Enhancing resources in indigenous schools improves academic scores and reduces bullying by encouraging student shifts from regular to indigenous schools. Policies targeting ethnic-based bullying improve academic performance directly and indirectly by fostering a conducive learning environment and encouraging students to choose more productive regular schools. Eliminating indigenous schools enhance academic achievement and reduce bullying for indigenous students attending regular schools, thanks to the influx of indigenous peers. Therefore, indigenous schools need additional resources to address minorities' education needs; otherwise, they risk being counterproductive.