The case definition for dengue hemorrhagic fever (DHF) requires fever, platelets < or = 100,000/mm3, any hemorrhagic manifestation, and plasma leakage evidenced by hemoconcentration > or = 20%, pleural or abdominal effusions, hypoproteinemia or hypoalbuminemia. We evaluated the specificity and yield of modified DHF case definitions and the recently proposed World Health Organization criteria for a provisional diagnosis of DHF, using a data base of laboratory-positive and laboratory-negative reports of hospitalizations for suspected dengue in Puerto Rico, 1994 to 1996. By design, all modifications had 100% sensitivity. More liberal criteria for plasma leakage were examined: 1) adding as evidence a single hematocrit > or = 50% (specificity 97.4%); 2) accepting hemoconcentration > or = 10% (specificity 90.1%); and 3) accepting either hematocrit > or = 50% or hemoconcentration > or = 10% (specificity 88.8%). The new DHF cases identified by these definitions (and percent laboratory positive) were 25 (100.0%), 95 (90.5%), and 107 (91.6%), respectively. In contrast, the provisional diagnosis of DHF (fever and hemorrhage, and one or more of platelets < or = 100,000/mm3, or hemoconcentration > or = 20%, or at least a rising hematocrit [redefined quantitatively as a 5% or greater relative change]) showed a specificity of 66.8%, and identified 318 new DHF cases, of which 282 (88.7%) were laboratory-positive. Very small changes in the criteria may result in a large number of new cases. The modification that accepted either hematocrit > or = 50% or hemoconcentration > or = 10% had acceptable specificity, while doubling the detection of DHF-compatible, laboratory-positive severe cases, but "provisional diagnosis" showed even lower specificity, and may produce inflated DHF incidence figures. Modified case definitions should be prospectively evaluated with patients in a health-care facility before they are recommended for widespread use.