“Rule out sepsis” (ROS) is a frequent neonatal diagnosis with few studies reporting complete populations. We prospectively identified all“sepsis work-ups” in 6 Kaiser Permanente hospitals from 10/95 through 11/96. We reviewed maternal and neonatal records, electronic laboratory data, and tracked 99.7% of infants to 1 week post discharge. During the study period, 19034 live births occurred and 2785 met entry criteria(birth weight ≥ 2kg and a sepsis evaluation consisting of a complete blood count &/or blood culture). Among the mothers, 30.3% had rupture of membranes > 18h before delivery, 14.1% antepartum temperature > 100.9°F, 20.1% a clinical diagnosis of chorioamnionitis, and 27.8% received parenteral antibiotics ≥ 4h before delivery. Of the 2785 infants, 76% were identified as being at risk at birth and 91% by 12 hours of age; 850 received systemic antibiotics; 205 were ventilated; 21 had a positive culture(12 group B streptococcus, 5 E. coli, 4 other); 29 had clinically confirmed infections with negative blood/cerebrospinal fluid cultures (4 deaths occurred in this group); and 76 were rehospitalized (2 with group B streptococcus sepsis). Babies were more likely to be infected if the maternal temperature was > 101.9°F (univariate odds ratio 3.4, 95% confidence interval, 1.4-8.1), if the mother had a confirmed perinatal genital pathogen culture (OR 5.4, 95% CI, 2.6-11.5), if the absolute neutrophil count (ANC) in the first 12 hours of life was < 9500 (OR 3.0, 95% CI 1.6-5.6), and if the immature to total (I:T) neutrophil ratio in the first 12 hours of life was ≥ 0.3 (OR 5.0, 95% CI 2.7-9.1). Asymptomatic babies were less likely to be infected (OR 0.33, 95% CI 0.17-0.64). Maternal temperature > 101.9°F strongly predicts neonatal infection but maternal temperature as a continuous variable is of limited utility, with an area under the receiver operator characteristic(ROC) curve of only 0.58. In contrast, the area under the ROC curve for the ANC was 0.71 and for the I:T ratio was 0.73. An evidence-based treatment/no treatment/length of treatment decision rule based on readily accessible predictors and validated on a separate dataset will be presented at the conference.