The present investigation was undertaken to standardize the early diagnosis of Gram-negative septicemia in burned children. Data were collected by means of a matrix which encompassed eight clinical variables routinely monitored by nursing personnel. These variables were evaluated according to their severity using a numerical scale of 0 to 3. A sepsis score was thus calculated for each of 243 burned patients, three times a day throughout their entire hospitalization. Eighty patients with suspiciously high scores (controls) were subjected to a battery of ten laboratory tests aimed at confirming the presence or absence of septicemia. During the 26 months of the study 16 patients (22 episodes) had clinical and laboratory evidence of Gram-negative septicemia. Multiple regression and discriminant analysis techniques were then used to develop statistical models for early diagnosis of septicemia. The two most practical and reliable of these are reported herein. Model I and II would have predicted the diagnosis of sepsis, 83% and 86% of the time, respectively, 1 day before the diagnosis was made using conventional methods. The false positive rates of Models I and II were 7% and 3%, respectively. On the basis of this information it seems possible and rewarding to utilize decision-making charts for monitoring and diagnosis of septicemia.