Clinical predictions alone are insufficiently accurate to identify patients with specific types of bloodstream infection; laboratory assays might improve such predictions. Therefore, we performed a prospective cohort study of 356 episodes of sepsis syndrome and did Limulus amebocyte lysate (LAL) assays for endotoxin. The main outcome measures were bacteremia and infection due to gram-negative organisms; other types of infection were secondary outcomes. Assays were defined as positive if the result was > or = 0.4 enzyme-linked immunosorbent assay units per milliliter. There were positive assays in 119 (33%) of 356 episodes. Assay positivity correlated with the presence of fungal bloodstream infection (P < .003) but correlated negatively with the presence of gram-negative organisms in the bloodstream (P = .04). A trend toward higher rates of mortality in the LAL assay-positive episodes was no longer present after adjusting for severity. Thus, results of LAL assay did not correlate with the presence of bacteremia due to gram-negative organisms or with mortality after adjusting for severity but did correlate with the presence of fungal bloodstream infection.