Determining sleep latency is one of the cornerstones of the interpretation of the multiple sleep latency test (MSLT). The purpose of this study was to compare various criteria used to determine sleep onset. We prospectively analyzed 100 consecutive MSLTs that were performed according to a standardized protocol. We scored each test using three separate sets of criteria for sleep onset: 1) one epoch of stage 1 sleep, 2) two consecutive epochs of stage 1 sleep, and 3) three consecutive epochs of stage 1 sleep. Each method yielded a mean sleep latency and a categorical classification of the record as normal if > 10 minutes, moderate if > or = 5 and < or = 10 minutes, and severe sleepiness if < 5 minutes. The ages of participants ranged from 4 to 78 years (mean 45.5). The averages of the mean sleep latencies across all three methods were: 6.2 minutes [standard deviation (SD) = 4.3] using one epoch, 7.2 minutes (SD = 4.7) using two epochs, and 7.5 minutes (SD = 4.9) using three epochs. Using the three categories of sleepiness, the implementation of the three-epoch criterion vs. the one-epoch criterion produced a change in category in 16 patients (16%). Five went from severe to moderate, 10 from moderate to normal, and 1 from severe to normal. As compared to using one epoch, using three produced an increase in mean sleep latency of at least 50% in 13 patients. The use of various criteria for sleep onset, especially criteria 1 and 3 above, produces differences in interpretation that are neither rare nor quantitatively negligible. Standardization of the methodology across centers would be desirable in clinical practice as well as for research protocols.