This article reports the use of a computer-based collaborative learning instruction designed to help students develop understanding of image formation by lenses. The study aims to investigate how students, working in dyads and mediated by multimedia computer-assisted learning (CAL) programs, construct shared knowledge and understanding. The subjects were a class of 36 Secondary 4 (Year 10) students working in dyads throughout the instruction. The instruction comprised three stages (namely, pre-test, computer-based activity, and posttest), during which students' within-dyad interactions were audio-recorded and transcribed for analysis. Three months after the instruction, some selected students were interviewed individually. The pre-test result showed that many students held the 'holistic conceptualizations' of image formation (rather than the physicists' point-to-point mapping model), which they applied to give alternative answer/explanations to the questions in the test. The post-test and interviews showed that overall, students improved their understanding of image formation although the improvement ranged widely. The rich qualitative data of peer interactions show that students experienced conflicts and co-construction that fostered their intensive engagement with tasks and with each other. The data also show the mediating role of the CAL programs and the teacher that helped students develop understanding.