Primary bloodstream infection (BSI) is a leading, preventable infectious complication in critically ill patients and has a negative impact on patients' outcome. Surveillance definitions for primary BSI distinguish those that are microbiologically documented from those that are not. The latter is known as clinical sepsis, but information on its epidemiologic importance is limited. We analyzed prospective on-site surveillance data of nosocomial infections in a medical intensive care unit. Of the 113 episodes of primary BSI, 33 (29%) were microbiologically documented. The overall BSI infection rate was 19.8 episodes per 1,000 central-line days (confidence interval [CI] 95%, 16.1 to 23.6); the rate fell to 5.8 (CI 3.8 to 7.8) when only microbiologically documented episodes were considered. Exposure to vascular devices was similar in patients with clinical sepsis and patients with microbiologically documented BSI. We conclude that laboratory-based surveillance alone will underestimate the incidence of primary BSI and thus jeopardize benchmarking.