The European Carotid Surgery Trial is a multicentre trial of carotid endarterectomy for patients who, after a carotid territory non-disabling ischaemic stroke, transient ischaemic attack, or retinal infarct, are found to have a stenotic lesion in the relevant (ipsilateral) carotid artery. Over the past 10 years 2518 patients have been randomised, and the mean follow-up is now almost 3 years among the 2200 thus far available for analysis of the incidence of strokes that lasted more than 7 days. For the patients with "moderate" (30-69%) stenosis on their prerandomisation angiogram the balance of surgical risk and eventual benefit remains uncertain, and full recruitment continues. For 374 patients with only "mild" (0-29%) stenosis there was little 3-year risk of ipsilateral ischaemic stroke, even in the absence of surgery, so any 3-year benefits of surgery were small, and were outweighed by its early risks. For 778 patients with "severe" (70-99%) stenosis, however, the risks of surgery were significantly outweighed by the later benefits: although 7.5% had a stroke (or died) within 30 days of surgery, during the next 3 years the risks of ipsilateral ischaemic stroke were (by life-table analysis) an extra 2.8% for surgery-allocated and 16.8% for control patients (a sixfold reduction, p less than 0.0001). There was also a small reduction in other strokes, and at 3 years the total risk of surgical death, surgical stroke, ipsilateral ischaemic stroke, or any other stroke was 12.3% for surgery and 21.9% for control (difference 9.6% SD 3.3, 2p less than 0.01). The main concern was to avoid disabling or fatal events, and, among severe stenosis patients, 3.7% had a disabling stroke (or died) within 30 days of surgery, an extra 1.1% surgery versus 8.4% control (p less than 0.0001) had a disabling or fatal ipsilateral ischaemic stroke by 3 years, and the total 3-year risk of any disabling or fatal stroke (or surgical death) was 6.0% surgery versus 11.0% control (overall difference 5.0% SD 2.3, 2p less than 0.05); but, for disabling or fatal stroke the control risks seemed to diminish after the first year, so delay of surgery by just a few months after clinical presentation might make this overall difference non-significant.