In sleep-disordered breathing, tracheal sounds disappear during apnea and vary cyclicly during hypopnea. We tried to detect these changes in tracheal sounds automatically with a personal computer, and we evaluated the diagnostic accuracy of this system for detecting sleep-disordered breathing. Polysomnography and tracheal sound recording were done in 33 subjects with possible sleep apnea/hypopnea syndrome. Eighteen had positive results defined as an apnea/hypopnea index greater than 15. Tracheal sounds were digitized and a personal computer was used to calculate short time power spectra (44-600 Hz) every 0.2 seconds by the fast Fourier transform. The moving averages (18 seconds) of the logarithms of the power spectra were calculated every 2 seconds. Transient falls in the moving averages of more than 12 dB were detected. Those that were lower than 5 dB above the level of background noise were classified as tracheal sound apneas. The number of falls of more than 12 dB and the number of tracheal sound apneas correlated strongly with the number of apneas plus hypopneas (r = 0.95) and with the number of apneas (r = 0.97), respectively. The sensitivity and specificity of tracheal sound analysis for the sleep apnea/hypopnea syndrome (as defined above) were 89% and 60%, respectively, when the criteria was more than 15 falls of more than 12 dB per hour. We conclude that tracheal sound analysis by this method is useful as a screening test for the sleep apnea/hypopnea syndrome.