When blood cultures turn positive, the attending physicians are usually notified immediately about Gram stain findings. However, information on the accuracy of Gram staining is very limited. We examined the accuracy of preliminary blood culture reports provided by a regional laboratory in an observational study including the years 1996, 2000 to 2001, and 2003. We used data from computer files and technicians' laboratory notes. The study was restricted to cultures with one morphological type. Using cultural identification as a reference, we estimated the sensitivity, specificity, and positive and negative predictive values (PPV and NPV, respectively) for the following defined morphological groups: gram-positive cocci in clusters, gram-positive cocci in chains or diplococci, gram-positive rods, gram-negative cocci, gram-negative rods, and yeasts. We further evaluated the Gram stain and wet mount findings for the most frequent bacterial species/groups. We obtained 5,893 positive blood cultures and the following results for the defined groups: sensitivity, range of 91.3 to 99.7%; specificity, 98.9 to 100%; PPV, 94.6 to 100%; and NPV, 99.0 to 100%. The sensitivity for the most frequent species was in the range 91.3 to 100%, with nonhemolytic streptococci having the lowest value (sensitivity, 91.3%; 95% confidence interval, 86.2 to 94.9%). Wet mount reports were less accurate (sensitivity of 30 to 70% for species with peritrichous motility), and Enterobacteriaceae (notably Salmonella spp.) accounted for 25% of the reports stating polar motility. In conclusion, we demonstrated a high accuracy of Gram stain reports, whereas wet mount microscopy was generally less accurate.