We simultaneously recorded tracheal sound and air flow from nine normal subjects (seven males and two females). Sound was picked up at the supra sternal notch with an air-coupled sensitive microphone held in a small airtight probe. Flow was measured at the mouth using a pneumotachograph Fleisch n degrees 2. Both sound and flow were directly digitized at a sampling rate of 5120 Hz and then divided in 128-sample blocks. For each sound block the frequency spectrum was computed using the fast Fourier transform. In order to evaluate instantaneous flow-rate from tracheal sounds we investigated eight methods divided in two groups of four. In the first group (i.e., reference curves methods), we assumed that a relationship existed between sound and flow and was thus reflected by the variations of certain parameters. We chose to use simple straightforward relationships, already known and published. We tested four different parameters. During a calibration phase, we built for each parameter P a reference curve representing the variations of P versus flow and being specific to each subject. Then, an unknown flow was evaluated in calculating P on a 128-sample block, and the reference curve gave the corresponding flow. In the second group, we made a hierarchial clustering analysis of sound spectra for revealing the frequency modifications, induced by the flow. We tested two kinds of spectra as well as two ways of associating a flow to a given cluster. This led us to four other methods for calculating the flow. All the eight methods but one gave a mean uncertainty in the measure of flow of about 15%.(ABSTRACT TRUNCATED AT 250 WORDS)