Magnetic resonance (MR) techniques have had a major impact in the last 10-15 years in understanding and managing multiple sclerosis. This review summarizes the current uses of MR in multiple sclerosis, based on the proceedings of a recent international workshop, under four headings: (i) technical issues; (ii) role in diagnosis; (iii) natural history studies in understanding the disease; (iv) application in clinical trials. The theory and methodology of relevant technical issues is outlined, in order to provide a framework with which to understand the potential and limitations of MR in addressing biological and clinical questions in multiple sclerosis. The principles underlying signal-to-noise and contrast-to-noise ratio are discussed, along with the techniques and clinical results for conventional and fast spin echo T2-weighted imaging, fluid-attenuated inversion recovery, detection of blood-brain barrier break down and hypointense lesions on T1-weighted images, magnetization transfer, T2 decay-curve analysis, MR spectroscopy, spinal cord imaging, diffusion imaging, and quantification of lesion load and atrophy. MRI has an extremely valuable role in confirming the clinical diagnosis of multiple sclerosis. T2-weighted brain imaging remains the standard diagnostic tool, but in some instances it is usefully complemented with gadolinium enhancement and spinal imaging. The caveat that the diagnosis of multiple sclerosis remains primarily a clinical one cannot be over-emphasized. Serial MRI studies have added much to our understanding of the natural history and pathophysiology of the disease. Blood-brain barrier breakdown is a consistent early feature of new lesion development in relapsing-remitting and secondary progressive multiple, sclerosis, and this usually correlates with active inflammation and myelin breakdown. A number of the acute MR changes are reversible, but chronic persistent abnormalities in a number of MR parameters, such as reduced N-acetyl aspartate, low magnetization transfer ratios, atrophy and T1-hypointensity, suggest the presence of demyelination and/or axonal degeneration in many chronic lesions. The presence and extent of T2-weighted MRI abnormalities at first presentation with a clinically isolated syndrome suggestive of demyelination strongly predicts the risk of developing clinically definite multiple sclerosis in the next few years. In established multiple sclerosis, however, the correlations between T2 abnormalities and disability are modest. This poor relationship partly relates to the discrepancy between lesion site and function in attempting to correlate locomotor disability with brain MRI findings. However, the correlations between brain lesion load and cognitive dysfunction in multiple sclerosis, whilst more evident, are still modest. A more important limitation is the low pathological specificity of abnormalities seen on T2-weighted images. Stronger correlations have been found between disability and new putative MR markers for demyelination and/or axonal degeneration. Serial studies using multiple MR techniques are now needed to further clarify pathophysiological mechanisms in multiple sclerosis. Serial MR has become an important tool in monitoring treatment efficacy. It provides data which can be readily analysed in a blinded fashion and which directly inspects the pathological evolution; it also enables a rapid and sensitive measure of treatment outcome in early relapsing-remitting and secondary progressive disease. Because of the modest clinical correlations it is, however, still appropriate that the definitive determinant of treatment efficacy remains a clinical one. Further work is needed to address issues of quality control in serial studies, statistical calculation of appropriate sample sizes, and optimization of the nature and frequency of MR outcomes measured.