Heart failure may lead to loss of complexity in physiologic dynamics. However, conventional entropy measures yield contradictory results. We applied a recently developed method, multiscale entropy (MSE) that quantifies the regularity of physiologic signals over multiple time scales. We tested the hypothesis that the rate at which entropy varies with time scale can discriminate between healthy and different classes of con- gestive heart failure (CHF) subjects. Material and Methods: We analyzed long-term Holter re- cordings from two group of subjects: 72 records from healthy subjects aged 5512 years and 43 from patients with CHF aged 5416. The CHF group included 12 subjects in NYHA classes I-II, and 31 subjects in classes III-IV. We calculated an entropy measure (SampEn) that applies to physiologic time series plotted as a function of time scale. Next, we computed the rate of change of entropy for scales smaller and larger than approximately 5 seconds and developed clustering criteria for the different groups. Results and Conclusions: Healthy and CHF subjects have different MSE curve profiles. Further, heart rate dynamics with CHF have significantly lower multiscale entropy values com- pared with healthy subjects. A 2-dimensional scatter plot based on the rate of variation of entropy with time scale showed robust separation of healthy and all CHF classes. These findings are consistent with the basic concept of information loss in cardio- vascular regulatory mechanisms in CHF and may be useful in clinical assessment.