To investigate biological and health services predictors of extreme neonatal hyperbilirubinemia in a health maintenance organization. Nested case-control study. Eleven Northern California Kaiser Permanente hospitals. The cohort consisted of 51,387 newborns born at 36 weeks or later weighing 2000 g or more. Cases were newborns with peak total serum bilirubin levels greater than or equal to 428 micromol/L (> or =25 mg/dL) (n = 73). Controls were a random sample of newborns from the cohort with peak bilirubin levels less than 428 micromol/L (<25 mg/dL) (n = 423). Review of medical records and telephone interviews. Early jaundice was most strongly associated with case status (odds ratio [OR] = 7.3). After excluding subjects with early jaundice, the strongest predictors of hyperbilirubinemia were family history of jaundice in a newborn (OR = 6.0), exclusive breastfeeding (OR = 5.7), bruising (OR = 4.0), Asian race (OR = 3.5), cephalhematoma (OR = 3.3), maternal age of 25 years or older (OR = 3.1), and lower gestational age (OR = 0.6/week). These variables identified 61% of newborns as very low risk (about 1/4200). However, the risk in the remaining 39% was still low (1/370). More cases (79%) than controls (59%) had newborn length-of-stay and follow-up consistent with the American Academy of Pediatrics guidelines, but phototherapy use within 8 hours of the time that the guidelines recommend was uncommon in both cases (26%) and controls (33%). There were no apparent cases of kernicterus. Prevention of extreme hyperbilirubinemia may require closer follow-up than is currently recommended by the American Academy of Pediatrics and more use of phototherapy than was observed in this study. To prevent extreme hyperbilirubinemia (> or =428 micromol/L [> or =25 mg/dL]) in 1 newborn, many newborns would need to receive these interventions.