Oftentimes the etiological diagnostic differentiation between viral and bacterial infections is problematic, while clinical management decisions need to be made promptly upon admission. Thus, alternative rapid and sensitive diagnostic approaches need to be developed. Polymorphonuclear leukocytes (PMNs) or phagocytes act as major players in the defense response of the host during an episode of infection, and thereby undergo functional changes that differ according to the infections. PMNs functional activity can be characterized by quantification and localization of respiratory burst production and assessed by chemiluminescent (CL) byproduct reaction. We have assessed the functional states of PMNs of patients with acute infections in a luminol-amplified whole blood system using the component CL approach. In this study, blood was drawn from 69 patients with fever (>38 °C), and diagnosed as mainly viral or bacterial infections in origin. Data mining algorithms (C4.5, Support Vector Machines (SVM) and Naïve Bayes) were used to induce classification models to distinguish between clinical groups. The model with the best predictive accuracy was induced using C4.5 algorithm, resulting in 94.7% accuracy on the training set and 88.9% accuracy on the testing set. The method demonstrated a high predictive diagnostic value and may assist the clinician one day in the distinction between viral and bacterial infections and the choice of proper medication.