We analyzed 24,584 cultures obtained from 3371 infants during a three-year period to determine how frequently body surface culture isolates matched those obtained subsequently from body fluids. The sensitivity, specificity, and predictive values of surface cultures were calculated daily for the two weeks up to and including the day that a body fluid culture was obtained. Isolates from cultures of material from the ear canal, nasopharynx, axilla, umbilicus, groin, rectum, stomach, and endotracheal tube were rarely the same as those recovered from blood or fluids of the cerebrospinal, joint, pleural, pericardial, and peritoneal spaces. With a frequency of sepsis of 3.3% among patients in our neonatal intensive care unit, the optimum sensitivity, specificity, and positive predictive values of surface cultures were 56%, 82%, and 7.5%, respectively. These values did not improve substantially for any specific times of surface cultures prior to sepsis, for anatomic sites cultured, or for pathogens recovered. We conclude that surface cultures are of limited value in predicting the etiology of sepsis in neonates.