The objective of this study was to define the limits of echocardiography and to evaluate thoracic spiral CT angiography (TSCTA) for the diagnosis of pulmonary embolism (PE). One hundred twelve consecutive patients, hospitalised for suspected PE, were included in this prospective study. All were investigated by pulmonary ventilation-perfusion scintigraphy (Sc) and 50 had a high probability of PE on this examination. Sc was normal in 22 patients. Forty patients were excluded because of an intermediate probability. In 50 patients with PE confirmed on Sc, transthoracic echocardiography (TTE) showed only indirect evidence of PE (intracavitary thrombus in 4% of cases). TSCTA demonstrated PE in 82% of cases and did not show any thrombus image when Sc was normal. Its negative predictive value was therefore 70% and its positive predictive value was 100%. Its sensitivity varied according to degree of perfusion defect (96% in the case of lobar lesion, 66% in the case of segmental lesion and 16% for a subsegmental lesion). Multidimensional transoesophageal echocardiography (TOE), performed in 37 of the 50 patients with PE, only revealed thrombi in the pulmonary tree in 3 patients (8%), all presenting severe PE. No thrombus was visualized on TOE in patients with non-serious PE. All thrombi observed on TOE were also demonstrated by TSCTA. In conclusion, TTE usually provides only indirect signs of PE. TOE has a poor diagnostic sensitivity for PE. TSCTA has a better sensitivity than TOE for the detection of thrombi in the pulmonary artery trunk and proximal centimetres of its two branches, but normal CT angiography cannot exclude a distal PE.