Thirty-five healthy, premature infants, ranging from 30-39 weeks postconceptional age, were observed continuously for 6 to 24 hr. Behavioral state and electroencephalographic patterns were coded for each minute. Using these data, three questions regarding coding of states of sleep were addressed: What is the concordance between behavioral codes and specific EEG patterns? Does the concordance between behavioral codes and EEG patterns change with postconceptional age? What range of error can be expected when observation periods shorter than 24-hr are used to estimate the daily distribution of quiet sleep (QS) and active sleep (AS)? With behavioral codes as the standard, concordances of EEG patterns for QS and AS were 72.5 and 92.1% respectively. With EEG patterns as the standard, behavioral codes for QS and AS agreed 83.0 and 88.9%. Agreement between behavioral codes and EEG patterns for QS increased with age. Finally, variation in estimates of the daily distribution of QS and AS decreased dramatically as the length of observation increased from 3 to 24 hr.