The objective of the present work was to develop automated methods for the compressed tracheal breathing sound analysis. Overnight tracheal breathing sound was recorded from ten apnoea patients. From each patient, three different types of tracheal sound deflection pattern, each of 10 min duration, were visually scored, viewing the compressed tracheal sound curve. Among them, high deflection patterns are of special interest due to the possible correlation with apnoea-hypopnoea sequences. Three methods were developed to detect patterns with high deflection, utilizing nonlinear filtering in local characterization of tracheal sounds. Method one comprises of local signal maximum, the second method of its local range, and the third of its relative range. The three methods provided 80% sensitivity with 57, 91 and 93% specificity, respectively. Method three provided an amplitude-independent approach. The nonlinear filtering based methods developed here offer effective means for analysing tracheal sounds of sleep-disordered breathing.