The study of the dynamics of non-linear systems allows the evaluation of the correlation dimension which, in turn, provides an estimate of the number of variables needed to model the process. In such a view, the correlation dimension was calculated for the profiles of the EEG slow-wave activity during sleep obtained from 7 young normal controls and in their corresponding artificial stochastic signals. It was possible to evaluate the complexity of all the real profiles which exhibited an average dimension of 3.76 (SD 0.331), but not that of the artificial control ones. This allows us to conclude that sleep regulation might be considered as a deterministic non-linear process and that the already proposed two-process model of sleep regulation needs to include additional variables with non-linear interactions.