OBJECTIVES: To assess the value of each of the Duke criteria for the diagnosis of infective endocarditis (IE). METHODS: Detailed review was done of charts of all cases discharged with the diagnosis of IE, and classification as 'definite', 'possible' and 'rejected' cases was made according to the Duke criteria. The diagnostic impact of each criterion was assessed by reclassifying each 'definite' case after subtraction of each individual criterion and by reclassifying each 'possible' and 'rejected' case after addition of each individual criterion. RESULTS: From 1983 to 1993, 179 cases were identified in the databases of two hospitals. When the Duke criteria were applied, 124 (6967%) were classed as 'definite', 43 (2466%) as 'possible' and 12 (763%) as 'rejected' cases. Of the 67 pathologically proven cases, 52 (78610%) were 'definite' cases when the criteria were applied before pathology. If the major microbiological criterion is subtracted, 53% (69%) of the 'definite' cases become 'possible' or 'rejected'. When the echocardiographic criterion is subtracted, 34% (68%) of the 'definite' cases become possible or rejected. Among minor criteria, fever and predisposition, contributing to the classification of respectively 31% (68%) and 27% (68%) of the 'definite' cases, were the most powerful. On the other hand, the minor microbiological criterion and immunologic phenomena were responsible for the classification of only 2% (62%) and 6% (64%) respectively, of the 'definite' cases. CONCLUSIONS: Depending on the criterion examined, 47-98% of the 'definite' cases of IE would remain 'definite' if this particular criterion were absent. The major microbiological criterion had the highest relative importance. In this retrospective study, in which only 32 (18%) patients had a transesophageal echocardiogram, the echocardiogram contributed to 15% (66%) of the 'definite' cases according to the major criterion and to 19% (66%) according to the minor criterion. This study illustrates that the degree of certainty of the diagnosis of IE often depends on the presence/absence of only one criterion.