The goal of this study was to develop and validate clinical prediction rules for bacteremia and subtypes of bacteremia in patients with sepsis syndrome. Thus, a prospective cohort study, including a stratified random sample of 1342 episodes of sepsis syndrome, was done in eight academic tertiary care hospitals. The derivation set included 881 episodes, and the validation set included 461. Main outcome measures were bacteremia caused by any organism, gram-negative rods, gram-positive cocci, and fungal bloodstream infection. The spread in probability between low- and high-risk groups in the derivation sets was from 14.5% to 60.6% for bacteremia of any type, from 9.8% to 32.8% for gram-positive bacteremia, from 5.3% to 41.9% for gram-negative bacteremia, and from 0.6% to 26.1% for fungemia. Because the model for gram-positive bacteremia performed poorly, a model predicting Staphylococcus aureus bacteremia was developed; it performed better, with a low- to high-risk spread of from 2.6% to 21.0%. The prediction models allow stratification of patients according to risk of bloodstream infections; their clinical utility remains to be demonstrated.