We investigated the time course of RR interval variability during exercise and subsequent 50 minutes of recovery in seven well-trained male cyclists who performed an exercise with 3 successive 8 min stages at 40 %, 70 % and 90 % of their maximal oxygen uptake. The goal of the study was to check whether the decrease in the amplitude of heart rate variability during heavy exercise was accompanied by changes in the chaotic structure of the fluctuations. Heart rate variability was analysed in the temporal and frequency domain using traditional tools and using non-linear methods (Largest Lyapunov Exponent, Detrended Fluctuation Analysis, Minimum Embedding Dimension). When compared to rest, variability at the heaviest exercise intensity was significantly lower (RR: 0.94 +/- 0.22 vs. 0.34 +/- 0.01 ms; SDRR: 0.11 +/- 0.04 vs. 0.01 +/- 0.00 ms) due to a decrease in both LF (2101 +/- 1450 vs. 0.14 +/- 0.09 ms (2) . Hz (-1)) and HF spectral energy (1148 +/- 1126 vs. 7.88 +/- 9.24 ms (2) . Hz (-1)). Non-linear analyses showed that heart rate variability remained chaotic whatever the exercise intensity (the largest Lyapunov exponent was positive at 90 % of the maximal oxygen uptake), with a fractal organisation that tended towards white noise (DFA value close to 0.5) during heavy exercise. During recovery, temporal and spectral variables came back to their rest values within about 30 minutes following an exponential pattern. Non-linear analyses revealed that heartbeat dynamics were disorganised at the beginning of recovery, and involved more regulating systems than at rest, even after 50 minutes of recovery. We concluded that, during heavy exercise, heart rate variability was mainly influenced by other factors than autonomous nervous system, and suggest that mechanical or neurological couplings between the cardiac, locomotor and respiratory systems could play an important part in the observed changes.