Daytime sleepiness is a common consequence of repeated arousal in obstructive sleep apnea (OSA). Arousal indices are sometimes used to make decisions on treatment, but there is no evidence that arousals are detected similarly even by experienced observers. Using the American Sleep Disorders Association (ASDA) definition of arousal in terms of the accompanying electroencephalogram (EEG) changes, we have quantified interobserver agreement for arousal scoring and identified factors affecting it. Ten patients with suspected OSA were studied; three representative EEG events during each of light, slow-wave, and rapid-eye-movement (REM) sleep were extracted from each record (90 events total) and evaluated by experts in 14 sleep laboratories. Observers differed (ANOVA, p < 0.001) in the number of events scored as arousal (totals ranged from 23 to 53 of the 90 events). Overall agreement was moderate (kappa = 0.47), but it was best for events during slow-wave sleep, moderate for REM, and poor for light sleep (kappa = 0.60, 0.52, and 0.28, respectively). Agreement was unrelated to arousal duration. We conclude that the ASDA definition of arousal is only moderately repeatable. Account should be taken of this variability when results from different centers are compared.