OBJECTIVE--To investigate the detection rate of cardiac sources of embolism by transoesophageal echocardiography in patients with focal cerebral ischaemic events and to relate the echocardiographic findings to other clinical findings. DESIGN--Prospective study with blinded analysis of the echocardiographic data and subsequent comparison with the other clinical findings. SETTING--Regional cardiothoracic unit based in a teaching hospital. PATIENTS--131 consecutive patients with focal ischaemic cerebral events (49 with a transient ischaemic attack, 77 with a cerebrovascular accident, and five with a retinal arterial embolus) referred for echocardiography. INTERVENTIONS--Full M mode, cross sectional, Doppler, and contrast echocardiography by both the precordial and transoesophageal techniques. RESULTS--Precordial echocardiography detected a cardiac abnormality in 72 patients. Transoesophageal echocardiography confirmed all the precordial findings (except left ventricular hypertrophy, which at present cannot be defined with this technique) and detected other abnormalities in a further 20 patients (18 with potential right-to-left shunts and two with valve vegetations). It also showed spontaneous contrast echoes in 27 of 28 patients with a large left atrium and showed atrial thrombus in three. Cardiac abnormalities were clinically detected in 53 patients, all of which were confirmed or documented by echocardiography. In the 78 patients with no clinically detectable cardiac abnormality six had mitral valve prolapse and one had a regional wall motion defect (identified by precordial echocardiography) and 17 had potential right-to-left shunts (11 of which were identified only by transoesophageal echocardiography). CONCLUSIONS--Transoesophageal echocardiography is more sensitive than precordial echocardiography in detecting potential sources of embolism in these patients. However, except for the detection of a potential right-to-left shunt, the yield in patients with no cardiac abnormality is low. Moreover, the abnormalities detected in those with previously detected cardiac disease merely confirm the clinical diagnosis. Patients with left atrial spontaneous contrast echoes may benefit from anticoagulation but this requires further study. Until more data are available on this feature and on the role of potential right-to-left shunts in this population, the contribution of echocardiography, precordial or transoesophageal, remains limited.