As the content knowledge of English language learners and immigrant students is assessed in their less proficient language, the validity and fairness of assessments related to those students have been widely discussed. However, such assessment issues have not received much attention in rapidly expanding bilingual education programmes, where students learn content subjects through a second/foreign/additional language (L2) and are often assessed in that language. This paper addresses this research gap by examining the mediating effect of language of testing on students' performance in content subject assessments. 387 question items in the Biology papers of a public examination in Hong Kong were analysed according to their cognitive and linguistic demands. The performance of 480 students (half taking the test in their first language and half in their L2) on the questions with different demands was then analysed with MANOVA and multiple regression. The results show that linguistic demands, especially productive linguistic demands (e.g. writing sentences and texts), had a more significant effect on the performance of students taking the test in their L2. These results underscore the role played by language in the academic achievement of bilingual education students, which in turn illuminate valid assessment design and pedagogy in these programmes.