Long-range correlations have been evidenced in a number of experiments, generally using over-learned and over-practiced tasks. We hypothesized that long-range correlation could represent the byproduct of learning. We analyzed the series of periods produced by a group of expert and a group of novices during prolonged trials on a ski-simulator. Results showed a very low variability in expert's series, as compared to novices. Fractal analyses showed that fluctuations were significantly more structured and correlated in experts. These results suggest that learning could be conceived as the progressive installation of complexity in the system.