The aim of this work is to develop an automatic system that can be used as an assistant tool for the detection and diagnosis of different kinds of sleep Apnea (Obstructive, Hypopnea and Central Apnea, respectively). Three nonlinear techniques were used for feature extraction: Central tendency measures (CTM), Lempel-Ziv complexity (LZC) and Approximate Entropy (ApEn) for oxygen saturation signals (SaO2). A statistical Comparison using (t – test) was performed for comparing the population mean of normal group with each of the Sleep Apnea groups for the nonlinear parameters. Three Hidden Markov Models (HMMs), based on Baum–Welch algorithm were proposed to estimate the optimal number of the parameters. The results have showed that the use of HMM and the nonlinear features gave promising results used for classifying Sleep Apnea diseases.