Objective: To compare the signals of pulse — diagnosis of healthy volunteer in different times in one day. Methods: After collecting the pulse waves of 42 healthy volunteers in 4 specific periods in one day, do pretreatment, parameter extracting basing on harmonic fitting, modeling, and identification by unsupervised learning and supervised learning with cross-validation step by step for analysis of the 4 groups. Finally, paired T-test and ANOVA were used in feature mining. Results: There are significant differences among the pulse-diagnosis signals of healthy volunteers in different times, and the accuracy rate is about 63%∼ 84%. Pulse rate, F1zuocun and F2zuocun, etc. are key features in classification. Conclusion: Signals of pulse-diagnosis in TCM of healthy human have circadian rhythmicity.