Bilingual programmes in which an L2 is used as the medium of instruction are becoming popular in different parts of the world, and content and language integrated learning (CLIL) is one variant of such programmes. Recent research on CLIL has gradually shifted from product-oriented (i.e. evaluating the effectiveness of CLIL in terms of language and content learning outcomes) to more process-oriented by exploring classroom interaction in detail to gain insight into effective pedagogy and teacher education. This paper compares two different settings of CLIL classrooms in Hong Kong (where CLIL is referred to as "English Medium Instruction"). In one setting, the Grade 10 teachers and students (aged 15) had experienced CLIL for the first three years of secondary education; in the second one, the Grade 10 teachers and students had just started to experience the CLIL approach. In our comprehensive analysis of 15 observed lessons, we found that the teacher--student interaction where CLIL had just started was more monologic: the teacher dominated the lesson talk, students had difficulties expressing their meaning in L2, question-and-answer sequences were constrained--a finding which would cause concern if replicated more generally. Our interpretation is that it takes time for both teachers and students to get used to integrated content and language learning, teachers need to develop skills to engage students in extended verbal exchanges, and students need to reach operational levels of L2 proficiency. The findings of this study yield significant implications for the successful implementation of CLIL in other contexts.