Detrended fluctuation analysis (DFA), a statistical technique that probes the fractal or scaling properties of time series, has been applied successfully to a variety of data (Goldberger et al. 2000; Peng et al. 1995). In relation to the human cardiac rhythm, R-R records during exercise have been reported to exhibit scaling properties qualitatively different from those obtained at rest (Karasik et al. 2002). Specifically, exercise records demonstrated anticorrelation at short timescales and correlation at intermediate scales, while with resting records the opposite crossover pattern was observed. The present study further investigated this question. Male human subjects (n=4) were fitted S810 Heart Rate Monitors (Polar Electro Oy, Professorintie 5, FIN-90440 Kempele). R-R intervals were recorded at rest, and when exercising at 50, 100 and 150W using a cycle ergometer at constant pedal velocity (50 rpm). Recording intervals varied from 20 min at rest to 10 min at 150W, and were chosen so as to yield usable records of approximately 1000 intervals, after discarding the first two minutes of each record. The DFA algorithm was obtained as a text file (Goldberger et al. 2000) and compiled using C++. Log-log plots were constructed of the first-order DFA fluctuation function against segment length. The gradient of these plots reflects the correlation properties of the R-R time series over different timescales. The plots showed no abrupt discontinuities of gradient either at rest or during exercise. Two subjects showed some indication of decreasing gradient as segment length increased, both at rest and during exercise, but for the remainder the plots were essentially linear. Correlation coefficients for all plots exceeded 0.95. Linear regressions were fitted and their parameters compared using repeated measures ANOVA. There was a significant effect of exercise level on the gradient of the plots (p<0.025). At 150W, the average gradient was 19% higher than at rest (1.089 v 0.914). There was also a significant effect of exercise level on intercept (p<0.005). The results of this study therefore do not support the contention of a qualitative difference in scaling behaviour between resting and exercise conditions. Correlation properties, as evidenced by the gradients of the plots both at rest and during exercise, were either unaffected by segment length, or demonstrated a modest decline at increasing timescales. Exercise, particularly at 150W, did however increase the correlation properties of the R-R time series, over all time scales, compared with the resting condition.