Using unique survey data on rural secondary school children, this paper evaluates the relative quality of Islamic secondary schools (i.e. madrasahs) in Bangladesh. Students attending registered madrasahs fare worse in maths and English than students attending non-madrasah schools. However, failure to account for non-random sorting over-estimates the negative influence of madrasahs on student achievement. Evidence on the magnitude of this bias is presented. Once selection effect is taken into account, madrasah disadvantage in English is small while that in maths becomes insignificant. Given the overall low level of achievement, this suggests that madrasah students perform just as poorly as those from non-madrasah schools in rural Bangladesh.