Although pulse oximetry has been used to determine the frequency and extent of hemoglobin desaturation during sleep, movement artifact can result in overestimation of desaturation unless valid desaturations can be identified accurately. Therefore, we determined the accuracy of pulmonologists' and technicians' interpretations of graphic displays of desaturation events, derived an objective method for interpreting such events, and validated the method on an independent data set. Eighty-seven randomly selected desaturation events were classified as valid (58) or artifactual (29) based on cardiorespiratory recordings (gold standard) that included pulse waveform and respiratory inductive plethysmography signals. Using oximetry recordings (test method), nine pediatric pulmonologists and three respiratory technicians ("readers") averaged 50 +/- 11% (SD) accuracy for event classification. A single variable, the pulse amplitude modulation range (PAMR) prior to desaturation, performed better in discriminating valid from artifactual events with 76% accuracy (P < 0.05). Following a seminar on oximetry and the use of the PAMR method, the readers' accuracy increased to 73 +/- 2%. In an independent set of 73 apparent desaturation events (74% valid, 26% artifactual), the PAMR method of assessing oximetry graphs yielded 82% accuracy; transcutaneous oxygen tension records confirmed a drop in oxygenation during 49 of 54 (89%) valid desaturation events. In conclusion, the most accurate method (91%) of assessing desaturation events requires recording of the pulse and respiratory waveforms. However, a practical, easy-to-use method of interpreting pulse oximetry recordings achieved 76-82% accuracy, which constitutes a significant improvement from previous subjective interpretations.