We compared six inflammatory mediators (C-reactive protein (CRP), interleukin-6 (IL-6), soluble tumour necrosis factor receptors (p55 and p75) and soluble adhesion molecules (ICAM-1, E-selectin)) as early diagnostic tests for neonatal sepsis, and studied the possible benefit of combining parameters. Blood samples were obtained from 166 consecutively admitted neonates, who were suspected to suffer from infection within the first week of life. Neonates were retrospectively classified as infected (sepsis, clinical sepsis or pneumonia), possibly infected, or non-infected. Twenty-four infected neonates had higher serum levels of all six mediators (all P < 0.05), and 18 possibly infected neonates had higher levels of CRP, IL-6, ICAM-1 and E-selectin (all P < 0.05), than neonates without infection (n = 124). Receiver operator characteristic plots showed that CRP was the single best diagnostic test. Multiple logistic regression modelling, including various combinations of two to six mediators, consistently showed that IL-6, in addition to CRP, predicted sepsis. With infected and possibly infected neonates as the reference standard, a combined test of CRP > or = 10 mg/l and/or IL-6 > or = 20 pg/ml had a sensitivity of 85%, specificity of 62%, and negative likelihood ratio of 0.24. Using infected neonates as reference standard alone, and including possibly infected as controls, sensitivity increased to 96%, whereas specificity decreased to 58%; a negative test result (CRP < 10 mg/l and IL-6 < 20 pg/ml) ruled out sepsis with high certainty (likelihood ratio = 0.07). CRP performed best as a diagnostic test for neonatal sepsis. Diagnostic accuracy was further improved by combining CRP and IL-6, whereas the other parameters (p55, p75, ICAM-1 and E-selectin) added no further diagnostic information.