The records of 153 patients who presented to an emergency department with diabetic ketoacidosis were reviewed to determine whether any admission evaluation laboratory data could serve as a predictor of occult or coexisting infection. Ten patients with admission radiographs already demonstrating active infection (pneumonia or tuberculosis) and two patients with wet gangrene of an extremity were not included in subsequent statistical analysis, as their infections were diagnosed on initial evaluation. Analysis of readily available admission variables revealed that when age, sex, temperature, glucose, serum bicarbonate, pH, total leukocyte count, and differential are subjected to univariate and multivariate discriminant analysis, only an elevation in band neutrophils reliably predicted infection. Approximately half of our patients with elevated band counts (10 or greater) had a coexisting occult infection. An elevated band count was predictive of an occult coexisting major infection with a sensitivity of 100% (19/19) and a specificity of 80% (98/122).