It is clear from this literature that fragmented sleep is less restorative than consolidated sleep, and leads to sleepiness-related daytime impairment. The optimal approach to the quantification of sleep fragmentation continues to be debated. Modest and erratic correlations between measures of sleepiness and traditional measures of EEG arousals have pushed investigators to try and find more sensitive measures of sleep fragmentation. Simply correlating various measures of sleep fragmentation with a measure of sleepiness has significant limitations. Since sleep fragmentation is not the only factor affecting daytime sleepiness, these correlations can be misleading. For example, a subject with severely fragmented sleep will show elevated sleepiness during the day. However, the overall correlation may be reduced because lack of fragmented sleep does not guarantee that the level of sleepiness will be low. Multivariate statistical modeling is needed to account for sources of variance simultaneously in the prediction of daytime sleepiness. In this way it may be possible to identify the optimal definition of sleep fragmentation. More studies are needed that evaluate "sub-cortical" arousals, EEG arousals, and daytime function simultaneously. Ideally, clarification of these measurement issues will lead to an improved understanding of sleep structure and the mechanism through which sleep fragmentation impacts daytime function.