Since its introduction, the multiple sleep latency test (MSLT) has played a major role in the diagnosis of narcolepsy. We assessed its diagnostic value in a series of 2,083 subjects of whom 170 (8.2%) were diagnosed with narcolepsy. The sensitivity of the combination of two or more sleep onset rapid eye movement (REM) periods (SOREMPs) with a mean sleep latency of < 5 minutes on an initial MSLT was 70% with a specificity of 97%, but 30% of all subjects with this combination of findings did not have narcolepsy. In some narcoleptics who had more than one MSLT, the proportion of naps with SOREMPs varied substantially from the initial MSLT to the follow-up test. The highest specificity (99.2%) and positive predictive value (PPV) (87%) for MSLT findings was obtained with the criteria of three or more SOREMPs combined with a mean sleep latency of < 5 minutes, but the sensitivity of this combination was only 46%. The combination of a SOREMP with a sleep latency < 10 minutes on polysomnography yielded a specificity (98.9%) and PPV (73%) almost equal to those obtained from combinations of MSLT findings, but the sensitivity was much lower. Our results suggest that the MSLT cannot be used in isolation to confirm or exclude narcolepsy, is indicated only in selected patients with excessive daytime sleepiness, and is most valuable when interpreted in conjunction with clinical findings.