A noninvasive means of detecting airway secretions in ventilator-dependent patients is desirable because endotracheal suctioning can result in life-threatening complications. In a patient who had copious secretions, we observed a sawtooth pattern on his flow-volume curve that disappeared after suctioning. Accordingly, we systematically examined the usefulness of a sawtooth pattern on flow-volume curves in detecting secretions in ventilator-dependent patients and compared its accuracy with clinical examination. Flow-volume curves were recorded in 50 ventilator-dependent patients over 1 min of spontaneous breathing. In 15 of these patients, clinical examination was performed by three clinicians to determine its accuracy in detection of secretions. Endotracheal suctioning was then performed to determine the presence or absence of secretions. Subsequently, the flow-volume curves of all 50 patients were played back on a video screen, and three observers, who were unaware of the results of suctioning, made a decision regarding the presence or absence of a sawtooth pattern. The sensitivity of the sawtooth pattern in detecting secretions ranged from 0.76 to 0.86, and specificity ranged from 0.86 to 0.90. The likelihood ratio of a positive test ranged from 5.55 to 7.97, whereas the likelihood ratio of a negative test ranged from 0.16 to 0.27. Interobserver agreement, assessed by the kappa statistic, was excellent: 0.76, 0.76, and 0.84. In the subgroup of patients evaluated by both clinical examination and flow-volume curve analysis, clinical examination was less accurate in 11 of the 15 patients. In conclusion, detection of a sawtooth pattern strongly suggests the presence of secretions, and the absence of this pattern suggests that secretions are unlikely to be present.