Active learning has been promoted as a key for national educational reform from elementary to higher education for the past few years. However, MEXT is replacing it with the phrase “independent, dialogical, and deep learning” in the new National Course of Study with a view to articulating more clearly the intention of the present reform. We can say the idea of deep active learning advocated by Matsushita et al. (2015) has had some impact on the attention to deep learning by MEXT. The purpose of this paper is to explore deep active learning in science education focusing on the practice and research of conceptual change, which have been trying to combine deep conceptual understanding with dialogue in the classroom. First, we reviewed Hypothesis-Experiment-Instruction as a representative practice of deep active learning in the history of Japanese science education and identified three problems. Then we showed that we can find some solution to them in recent conceptual change research. Finally, by examining several assessment methods related to conceptual change, we proposed that Toulmin’s argument model would be effective to guide and assess the product as well as the process of conceptual change.