It is recommended that blood cultures be performed on all patients admitted to the hospital with pneumonia. Questions regarding the cost-effectiveness of this practice have emerged. We used data on 13,043 Medicare patients hospitalized with pneumonia to determine predictors of bacteremia. Predictors included recent antibiotic treatment, liver disease, and three vital-sign and three laboratory abnormalities. Patients were stratified into three groups on the basis of the likelihood of bacteremia. We then created a decision support tool that recommends performing no blood cultures on patients with low likelihood of bacteremia, one blood culture on patients with moderate likelihood of bacteremia, and two blood cultures on patients with higher likelihood of bacteremia. This tool was then applied to a validation cohort of 12,771 patients with pneumonia. Use of the decision support tool would result in 38% fewer blood cultures being performed when compared with the standard practice of performing two blood cultures for each patient and identified 88 to 89% of patients with bacteremia. A simplified tool performed similarly overall but was less sensitive than was the first tool among pneumonia severity index Class V patients. These tools may allow clinicians to target patients with pneumonia in whom blood cultures are most likely to yield a pathogen.