There is a paucity of data regarding predictive values and likelihood ratios of clinical signs for the diagnosis of late onset neonatal septicemia. This study aimed to determine these parameters in a prospective fashion, deriving a score by combining the most useful signs and determining the diagnostic utility of the score. All neonates admitted to a neonatal unit over a 1-year period were monitored for the occurrence of 16 pre-defined clinical signs. Symptomatic episodes (105 episodes in 80 neonates) were investigated for sepsis, and diagnosed as definite sepsis (n = 30), most probable sepsis (n = 17), and no sepsis (n = 58). Seven clinical signs (grunting, abdominal distension, increased pre-feed aspirates, tachycardia, hyperthermia, chest retractions, and lethargy) had positive likelihood ratios (PLR) greater than 1, and were combined to make a composite score. When a weighted clinical score (WCS) was used to diagnose definite sepsis, a cut-off score of 2 gave the best positive predictive value (PPV) and PLR (52 per cent and 2.65, respectively), and a cut-off score of 1 gave the best negative predictive value (NPV) and negative likelihood ratio (NLR) (85 per cent and 0.44, respectively). A cut-off score of 2 had a PPV of 65 per cent for definite and/or probable sepsis. In conclusion, physicians who attempt to make a diagnosis of neonatal sepsis on purely clinical grounds can use a seven-item weighted clinical score.