Nuclear magnetic resonance (NMR) imaging is an established diagnostic medium to diagnose multiple sclerosis (MS). In clinically stable MS patients, NMR detects silent disease activity, which is the reason why it is being used to monitor treatment trials, in which it serves as a secondary outcome parameter. The absence of a clear correlation with clinical disability, the so-called 'clinico-radiological' paradox, and the poor predictive value of NMR prohibit the use of NMR as a primary outcome parameter in clinical trials. This is--among others--a result of the limited histopathological specificity of conventional, or 'T2-weighted' imaging, the most commonly used NMR technique. In this paper we review additional NMR techniques with higher tissue specificity, most of which show marked heterogeneity within NMR-visible lesions, reflecting histopathological heterogeneity. Gadolinium enhancement identifies the early inflammatory phase of lesion development, with active phagocytosis by macrophages. Persistently hypointense lesions on T1-weighted images ('black holes') relate to axonal loss and matrix destruction, and show a better correlation with clinical disability. Marked prolongation of T1 relaxation time correlates with enlargement of the extracellular space, which occurs as a result of axonal loss or oedema. Axonal viability can also be measured using the concentration of N-acetyl aspartate (NAA) using NMR spectroscopy; this technique is also capable of showing lactate and mobile lipids in lesions with active macrophages. The multi-exponential behaviour of T2 relaxation time in brain white matter provides a tool to monitor the myelin water component in MS lesions (short T2 component) as well as the expansion of the extracellular space (long T2 component). Chemical exchange with macromolecules (e.g. myelin) can be measured using magnetization transfer imaging, and correlates with demyelination, axonal loss and matrix destruction. Increased water diffusion has been found in MS lesions (relating to oedema and an expanded extracellular space) and a loss of anisotropy may indicate a loss of fibre orientation (compatible with demyelination). Apart from the histopathological heterogeneity within focal MS lesions, the normal-appearing white matter shows definite abnormalities with all quantifiable NMR techniques. A decrease in the concentration of NAA, decreased magnetization transfer values and prolonged T1 relaxation time values are probably all related to microscopic abnormalities, including axonal damage. This 'invisible' lesion load may constitute a significant proportion of the total lesion load but is not visible on conventional NMR. Similarly, mechanisms for clinical recovery exist, which are not distinguished using MR imaging. Therefore, it is highly unlikely that the clinico-radiological paradox will ever be solved completely. However, NMR provides an opportunity to sequentially measure tissue changes in vivo. Using MR parameters with (presumed) histopathological specificity, the development of (irreversible) tissue damage can be monitored, which perhaps allows the identification of factors that determine lesional outcome in MS. Since the absence of severe tissue destruction is prognostically favourable, NMR monitoring of the extent to which such changes can be prevented by treatment will ultimately benefit the selection of future treatment strategies.