Objectives-The accuracy of magnetic resonance angiography (MRA) was determined in patients with recently symptomatic tight (80%-99%) carotid stenosis (on Doppler ultrasound), and the effect of stenosis severity on the accuracy and interobserver variability of MRA was studied. Methods-Forty four consecutive patients undergoing intra-arterial angiography (IAA) before carotid endarterectomy were prospectively studied, in two centres with identical MR scanners and sequences. All patients had undergone Doppler ultrasound, showing a 70% or worse carotid stenosis on the symptomatic side. MRA and IAA were done during the same admission. The MRA films were each independently and blindly read for percentage stenosis (signal gap if present) by four observers. The IA angiograms were read separately by one observer, blind to symptoms, and Doppler and MRA results. Results-Signal gaps on MRA were seen in stenoses ranging from 67% to 99% on intraarterial angiography. Magnetic resonance angiograms consistently overestimated the percentage stenosis according to intraarterial angiography. Clinically significant misclassification of stenosis occurred according to MRA in 7% of patients, and was more frequent as carotid stenosis increased. Conclusion-Significant diagnostic errors occur with MRA in patients with tight carotid stenosis. Any morbidity occurring as a result of misclassification by MRA is likely to be offset by the avoidance of complications; however, this could only be determined with certainty in a randomised controlled trial.