In this study, we applied generalized autoregressive conditional heteroskedasticity (GARCH) model to conditional fluctuation characteristics of heart rate variation (HRV) series (congestive heart failure, Normal), with all the data from PhysioNet ECG database. Research results proved the existence of condition fluctuation characteristic in the series of changing rate of HRV. In the GARCH model family, threshold GARCH (1,1)(TGARCH (1,1)) model performs best in fitting changing rate of HRV. Although the structure of ARCH (1) model is simple, its error is the closest to that of TGARCH (1, 1) model. The results also showed that the difference was obvious between disease group and normal group. All these results provide a new method to the research and clinical application of HRV.