The recent use of spectral analysis of the R-R interval variability to assess the autonomic drive during exercise has produced inconsistent results. The purpose of this study was to assess whether the spectral components of the R-R interval variability reflect different mechanisms at rest and during exercise. Autoregressive spectral analysis of the electrocardiographic and breathing signals was performed in 11 healthy young men at rest and during incremental cycle ergometry. The amplitude of respiratory sinus arrhythmia, i.e. the absolute power of the high frequency spectral component, fell at the onset of exercise, consistent with a reduction in cardiac vagal activity. Conversely, the normalized power of the high frequency component, (i.e. the percentage of R-R interval variance due to the respiratory sinus arrhythmia) increased with increasing work rates. The low frequency spectral component of the R-R interval variability, which when expressed in normalized units is believed to reflect cardiac sympathetic activity, was no longer detectable in severe exercise when the adrenergic drive is known to be elevated. In conclusion, autoregressive spectral analysis of the R-R interval variability does not adequately reflect the autonomic changes that occur during incremental exercise. In particular, the evidence indicates that as the cardiac vagal tone falls with increasing levels of exercise, a greater percentage of the residual power of the high frequency component may be due to non-neural mechanisms.