We analyzed surgically resected endocardial specimens from 49 patients by broad-range PCR. PCR results were compared with (1) results of previous blood cultures, (2) results of culture and Gram staining of resected specimens, and (3) clinical data (Duke criteria). Molecular analyses of resected specimens and previous blood cultures showed good overall agreement. However, in 18% of patients with sterile blood cultures, bacterial DNA was found in the resected materials. When data from patients with definite or rejected cases of infective endocarditis (IE) were included, the sensitivity, specificity, and positive and negative predictive values of broad-range PCR were 82.6%, 100%, 100%, and 76.5%, respectively, overall, and 94.1%, 100%, 100%, and 90%, for cases of native valve endocarditis. The sensitivity, specificity, and positive and negative predictive values of culture of resected specimens from patients with native valve endocarditis were 17.6%, 88.9%, 75%, and 36.4%. We recommend broad-range PCR of surgically resected endocardial material in cases of possible IE, in cases of suspected IE in which blood cultures are sterile, and in cases in which organisms grow in blood cultures but only Duke minor criteria are met. We propose to add molecular techniques to the pathologic criteria of the Duke classification scheme.