Our purpose was to assess the reproducibility of and differences between the most commonly used methods for assessing carotid artery stenosis using magnetic resonance angiography (MRA). We studied 55 patients who underwent axial three-dimensional time-of-flight MRA (1.5 T). Quantitative caliper measurements were performed from maximum intensity projection (MIP) and multiple planar reconstruction (MPR) images, according to the criteria of the North American Symptomatic Carotid Endarterectomy Trial (NASCET) and European Carotid Surgery Trial (ECST). The measurements were compared to each other and to visual interpretation, using conventional angiography as the reference. The measured percentage stenoses were higher on MRA than on digital subtraction angiography (DSA) using both NASCET (mean difference 1.9-3.0%) and ECST (6.3-6.7%) criteria. The kappa coefficients for the agreement between DSA and MRA were higher using the NASCET (0.61-0.76) than the ECST criteria (0.52-0.65). No statistically significant differences were found between measurements from MIP and MPR images. The ECST measurement criteria gave significantly higher percentage stenoses than the NASCET criteria (P < 0.001), this difference being more prominent on MRA (mean difference in diameter stenosis percentage 14.3-16.4%) than on DSA (7.6-11.2%) and most important with mild stenoses. The difference between visual interpretation and quantitative measurements on MRA was significant (P = 0.01-0.001). There were no statistically significant interobserver differences in the MRA film readings, either in visually estimated degrees of stenosis or stenosis measurements. Thus, the different criteria of the two multicentre trials led to significantly different results, especially in the assessment of mild stenosis, and these differences are more important with MRA than with DSA. Differences between the imaging modalities or the reconstruction programs seem less important.