Thirty virologically confirmed cases of dengue infection with a fatal outcome were studied clinically in Jakarta, Indonesia, from 1975 to 1978. All 4 dengue virus serotypes were isolated from fatal cases, but dengue type 3 was responsible for 21 (70%) of these isolates, compared to only 47% of isolates from all cases of dengue infection. The majority (60%) of these 30 cases were males in the 5-9-year age group. Nonspecific signs and symptoms in the fatal cases were no different from those in patients who survived dengue infection, but 70% of the patients with fatal outcome had one or more signs of encephalitis, primarily convulsions and somnolence; 3 of them developed spastic tetraparesis before death and 2 died of an illness clinically compatible with viral encephalitis. Other unexpected observations were that only 63% of the patients had classical dengue shock syndrome with haemoconcentration, thrombocytopenia and shock. A high percentage (80%) had gastrointestinal haemorrhage, and in 9 patients (30%) this was severe enough to cause shock and death. In these 9 cases, the gastrointestinal haemorrhage and haematemesis began before the onset of shock and there was no evidence of haemoconcentration or pleural effusion at any time during hospitalization. According to certain widely accepted criteria, these patients would not be diagnosed as dengue haemorrhagic fever (DHF). But as they made up nearly one-third of the confirmed fatal dengue infections in this study and had massive gastrointestinal haemorrhages with thrombocytopenia, the definition of DHF should be changed to include this type of patient. It is proposed that the disease should be more realistically classified as dengue fever with or without haemorrhage and dengue shock syndrome.