A novel multiplex real-time PCR for bloodstream infections (BSI-PCR) detects pathogens directly in blood. This study aimed at determining the positive predictive value (PPV) of BSI-PCR in critically ill patients with sepsis. We included consecutive patients with presumed sepsis upon admission to the intensive care unit (ICU). The multiplexed BSI-PCR included 17 individual PCRs for a broad panel of species- and genus-specific DNA targets. BSI-PCR results were compared with a reference diagnosis for which plausibility of infection and causative pathogen(s) had been prospectively assessed by trained observers, based on available clinical and microbiological evidence. PPV and false positive proportion (FPP) were calculated. Clinical plausibility of discordant positive results was adjudicated by an expert panel. Among 325 patients, infection likelihood was categorized as confirmed, uncertain, and ruled out in 210 (65%), 88 (27%), and 27 (8%) subjects, respectively. BSI-PCR identified one or more microorganisms in 169 (52%) patients, of whom 104 (61%) had at least one detection in accordance with the reference diagnosis. Discordant positive PCR results were observed in 95 patients, including 30 subjects categorized as having an “unknown” pathogen. Based on 5525 individual PCRs yielding 295 positive results, PPV was 167/295 (57%) and FPP was 128/5525 (2%). Expert adjudication of the 128 discordant PCR findings resulted in an adjusted PPV of 68% and FPP of 2%. BSI-PCR was all-negative in 156 patients, including 79 (51%) patients in whom infection was considered ruled out. BSI-PCR may complement conventional cultures and expedite the microbiological diagnosis of sepsis in ICU patients, but improvements in positive predictive value of the test are warranted before its implementation in clinical practice can be considered.