Heparin and warfarin were first widely used to treat pulmonary embolism (PE) in the early 1940s. Early acceptance of these drugs was widespread and, in retrospect, rather uncritical. In the face of enthusiastic reports it became very difficult to mount properly controlled trials. Other treatments which evolved in a similar way, and have also been in widespread use, have proved ineffective when formally assessed. It is the purpose of this paper to ask the question: 'How do we know that anticoagulants are of benefit in the treatment of PET Review of the literature Barritt & Jordan (1960) A widely quoted paper on the treatment of PE appeared in the Lancet in 1960. Barritt and Jordan embarked on a trial of anticoagulants (heparin and nicoumalone) versus no treatment, but abandoned it when they found that 5of their original 19 untreated patients had died; none of their 16 treated patients had died. As this is the only prospective controlled trial showing benefits from treatment, it will be reviewed in detail. Patients were referred into the trial by doctors other than the authors, so that selection was not random. Furthermore, the trial was not conducted in a double-blind fashion. No information is provided regarding the comparability of the treated and untreated patients. The incidence of massive pulmonary infarction among the patients who died (40%) is unusually high. Most other studies quote a 10% incidence of this complication (Moser 1977). The criteria used to make a diagnosis of PE arouse further doubts. Symptoms thought suggestive of PE were 'central chest pain, haemoptysis and faintness'. In a larger, more recent series using angiography to establish the diagnosis, these symptoms were present in fewer than 23% of PE patients (Urokinase PE Trial 1973). Two constellations of signs were supposed to be diagnostic of PE: (l) 'right heart failure: a fall in blood pressure, rise in JVP, together with changes in the ECG'; (2) 'pulmonary infarction: haemoptysis, fever, pleural friction, loss of resonance at the lung bases, rales, and in these patients it was usual to find changes in the bedside chest radiograph'. ECG changes are now thought to be nonspecific (Bell 1977), and from the clinical signs described for pulmonary infarction, one wonders how many patients in this series were placed on anticoagulants for pneumonia. Regarding non-fatal recurrences, no information is given, other than that 5 occurred in the untreated group. Considering the 5 deaths in the untreated group, one was due to cerebral infarction. This patient also had a PE, but so do between 20% and 60% of all persons who come to postmortem (Freiman et al. 1965). Four deaths were said to be due to PE : 2 of these patients died from sepsis (pneumonia with lung cavitation), one patient died 48 hours after the lodging of the first embolus, and one died immediately after a second embolization. The high incidence of lung cavitation among the deaths attributed to PE is striking. Sevitt & Gallagher (1961) found only 5% of lung cavitation among their PE fatalit...