Nonlinear dynamic analysis was used to examine the changes of complexity of heart rate variability (HRV) under rest and exercise. Nine healthy male subjects participated in the test. Each subject had a rest in supine position for 10 min, and then exercised on a cycle ergometer (10 min at 50 W and 10 min at 100 W). The results showed that R-R interval and its standard deviation reduced, and at the same time the values of nonlinear dynamics indices (complexity, entropy, correlation dimension, capacity dimension) also decreased during exercise. Moreover, at 100W work load the decrease of these indices was greater than that of 50W. It shows that physical exercise may reduce the complexity of HRV, and nonlinear dynamic indices could be used in distinguishing the complexity of heart rate variability under different physiological conditions.