The use of magnetic resonance imaging as a surrogate outcome measure in clinical trials, or even as a prognosticator in the assessment of the natural evolution, assumes a close relationship between extent and rate of development of magnetic resonance imaging abnormalities with the clinical status and rate of development of disability. While it may seem obvious that patients who develop new lesions are worse off than those without new lesions, the association between clinical findings and radiological extent of involvement is generally poor. In this review, various confounders are discussed, including inappropriate clinical rating, lack of histopathological specificity (especially for axonal loss), neglect of spinal cord involvement, underestimation of damage to the normal appearing brain tissue (both white and gray matter), and masking effects of cortical adaptation. It is concluded that much progression has been made in magnetic resonance techniques so that the clinico-radiological dissociation has indeed proved to be a paradox. Thus, the relevance of normal appearing brain tissue damage, residual brain volume, spinal cord damage and cerebral plasticity had to be reiterated. The increased awareness of the subtle interplay between these dimensions should be kept in mind when magnetic resonance is used as a surrogate outcome measure. This corroborates with conventional wisdom that one should not rely on a single magnetic resonance measure, but take full advantage of the fact that magnetic resonance is able to provide multidimensional information.