Methods based on non-linear heart rate (HR) dynamics have been suggested to probe features in HR behaviour that are not easily detected by the traditional HR variability indices. This study tested the hypothesis that analysis of correlation properties of R-R intervals provides useful information on HR fluctuation during exercise. High- (HF) and low-frequency (LF) spectral components and a short-term scaling exponent (alpha1) of HR variability, were analysed for nine healthy subjects at rest, during incremental and steady-state exercise, during atropine infusion and during incremental exercise after atropine administration. During the incremental exercise test alpha1 increased from rest to an intensity level of approximately 40% of VO2max (from 1.07+/-0.24 to 1.50+/-0.25, P<0.001) and thereafter decreased linearly until the end of exercise (from 1.50+/-0.25 to 0.38 +/- 0.10, P<0.001). Atropine infusion increased the scaling exponent alpha1 value from 0.91+/-0.23 to 1.37+/-0.31 (P<0.001). During exercise after atropine infusion, a linear reduction was observed in the scaling exponent alpha1 from 1.37+/-0.23 to 0.25+/-0.08 (P<0.001). Analogous changes in alpha1 were seen during long-term steady-state exercise compared to incremental exercise. Conventional HR variability indices did not show any significant changes during exercise at high exercise intensity levels. alpha1 correlated with the LF/HF ratio at rest (r=0.90, P<0.001), but the correlation was weaker after atropine (r=0.71, P<0.05) and during exercise (e.g. r=0.33, P=NS at the level of 40% of VO2max). In conclusion, incremental exercise test until exhaustion results in bidirectional changes in correlation properties of R-R interval dynamics. These changes can be explained by the intensity of vagal and sympathetic input to the sinus node during the different intensity levels of exercise. Changes in alpha1 values can be detected also in high intensity levels, when the conventional measures of HR variability can not be applied.