This paper focuses on a circadian rhythm (which is approximately 24-hour periodic rhythm) and proposes its estimation method based on the heartrate data obtained from a non-contact mattress sensor during sleep, instead of invasive and contacted devices which are hard to be used in a daily life. For this purpose, this paper proposes the circadian rhythm estimation method based on knowledge distillation without the daytime heartrate data, which executes as follows: (i) the full-day heartrate data is estimated by Fourier series regression using “the several days” of the nighttime heartrate; and (ii) the student model in the knowledge distillation is trained not only to estimate the circadian rhythm by “the estimated full-day heartrate,” but also to be closed to the teacher model that can estimate more accurate circadian rhythm trained by “the full-day core body temperature” obtained by other devices as the correct data. To investigate the effectiveness of the proposed method, the human subject experiment of the six days was conducted and the following implications have been revealed: (1) the averaged bottom peak (i.e., the minimum point) time of the circadian rhythm estimated by the proposed method is closer to the correct time than that by the proposed method without knowledge distillation; (2) the averaged top peak (i.e., the maximum point) time of the circadian rhythm estimated by the proposed method is mostly the same as that of the proposed method without knowledge distillation, but the valiance of the top peak time of the circadian rhythm estimated by the proposed method is smaller than that by the proposed method without knowledge distillation.