In the period April 1981 through June 1982, there was an unusual increase in the number of deaths and cardiopulmonary arrests in the pediatric intensive care unit at a large medical center hospital in San Antonio, Texas. During this period, 34 of 42 deaths (81 per cent) occurred in the evening work shift, as compared with 36 of 106 (34 per cent) during the previous four years (P less than 0.0001). Reviews of records of patients revealed no association between death in this 15-month epidemic period and a variety of demographic, historical, medical, and admission characteristics; medical or surgical procedures; or the severity of illness. The findings of a blinded clinical consultant support the conclusion that the increase in deaths and cardiopulmonary arrests could not be explained on the basis of the clinical status of the patients, and the consultant concluded that during the epidemic period, there were more deaths and cardiopulmonary arrests that were either unexpected in timing or inconsistent with the previous clinical course. The presence of one nurse was associated with the increased numbers of deaths (relative risk, 10.7; 95 per cent confidence limits, 6.4 to 17.9), cardiopulmonary arrests (relative risk = 25.5, confidence limits = 16.8 to 38.6), and unexpected clinical events, (relative risk, infinite; confidence limits, 33.7 to infinity). Although the cause of the epidemic remains unclear, the findings illustrate that surveillance of deaths and cardiopulmonary resuscitation may allow early recognition of similar problems in other hospitals.