In lectures with presentation slides such as an e-learning lecture on video, it is important for lecturers to control their non-verbal behavior involving gaze, gesture, and paralanguage. However, it is not so easy even for well-experienced lecturers to properly use non-verbal behavior in their lecture to promote learners' understanding. This paper proposes robot lecture, in which a robot substitutes for human lecturers, and reconstructs their non-verbal behavior to enhance their lecture. Towards such reconstruction, we have designed a model of non-verbal behavior in lecture. This paper also demonstrates a robot lecture system that appropriately reproduces non-verbal behavior of human lecturers with reconstructed one. In addition, this paper reports a case study involving 36 participants with the system, whose purpose was to ascertain whether robot lecture with reconstruction could be more effective for controlling learners' attention and more beneficial for understanding the lecture contents than video lecture by human and robot lecture with simple reproduction. The results of the case study with the system suggest the effect of promoting learners' understanding of lecture contents, the necessity of reconstructing non-verbal behavior, and the validity of the non-verbal behavior model.