Consciousness has not yet been satisfactorily defined because of its puzzling nature which involve the perception of the environment (perceptual awareness) and of the self (self-awareness). Current available methods fail in establishing prognosis in patients with vegetative state (VS): to our mind, this failure stems from the heterogeneous localization of brain damages causing VS and from available approaches tending to investigate self-awareness separately from perceptual awareness, whereas consciousness should be explored as a single and indivisible whole. Moving from the assumption that consciousness depends on the normal activity of wide neural networks, that may be regarded as complex systems whose outputs show a nonlinear behaviour, we propose a nonlinear approach applied to electroencephalographic (EEG) signal, aimed at exploring residual neural networks complexity in patients with VS. For this objective the EEG recording of 10 patients previously admitted to our department were retrospectively analyzed and compared with those of ten matched healthy control subjects. Approximate Entropy (ApEn) was calculated from the average values of time series with fixed input variables. Mean ApEn values were lower in patients than in controls (t18 18 = 12.3, p < 0.001). ApEn is able to discriminate patients from controls thus supporting the hypothesis about a decreased neural networks complexity in VS.