The present project proposes to integrate disperse knowledge and techniques in automatic sleep analysis, into conceptual models that lead to a deeper knowledge of the underlying mechanisms, and have discrimination power. The basic hypothesis is that sleep is a highly structured process, both in macro and microstrutural views, and thus can be modeled in terms of its instantaneous activity and its temporal dynamic. The proposed methodology for the analysis of the electroencephalogram (EEG) consists of a hierarchical model that explores sophisticated processing, modeling, analysis and classification techniques. The innovative character of this project lies on the methodologies and sophisticated techniques for processing, modeling, representation and classification of electroencephalographic signals and on the global coherent view of EEG. The proposed hierarchical/syntatic approach leads to sleep modeling unifying micro and macrostructure in a single representation. This approach introduces objective quantitative methods of analysis and classification of the EEG, providing additionally a significant data reduction. In pratical terms, this representation enables the characterization of sleep in terms of new micro-states, a higher discrimination capacity between populations and the possibility to predict future states.