Recent MRI studies in multiple sclerosis have highlighted the potential importance of spinal cord atrophy (implicating axonal loss) in the development of disability. However, the techniques applied in these initial studies have poor reproducibility which limits their application in the serial monitoring of patients. The aim of this study was to develop a highly reproducible and accurate method for the quantification of atrophy. The technique we describe demonstrates an intra-observer coefficient of variation (scan-rescan) of only 0.8%. When applied to 60 patients with clinically definite multiple sclerosis there was a strong correlation between spinal cord area and disability measured by Kurtzke's Expanded Disability Status Scale (EDSS) (r = -0.7, P < 0.001). The correlation was graded providing evidence for a causal connection. At levels 3 and 8 of the EDSS we observed a reduction in average cord area of 12 and 35%, respectively. Given its reproducibility, the magnitude of the change detected and the strong correlation with disability, this new technique should prove to be a sensitive measure of progressive neurological deterioration and could be readily incorporated into imaging protocols aimed at monitoring therapy.