Tumour necrosis factor mediates chronic inflammatory pathologies including those affecting the central nervous system, but non-selective tumour necrosis factor inhibitors exacerbate multiple sclerosis. In addition, TNF receptor SF1A, which encodes one of the tumour necrosis factor receptors, has recently been identified as a multiple sclerosis susceptibility locus in genome-wide association studies in large patient cohorts. These clinical data have emphasized the need for a better understanding of the beneficial effects of tumour necrosis factor during central nervous system inflammation. In this study, we present evidence that the soluble and transmembrane forms of tumour necrosis factor exert opposing deleterious and beneficial effects, respectively, in a multiple sclerosis model. We compared the effects, in experimental autoimmune encephalomyelitis, of selectively inhibiting soluble tumour necrosis factor, and of both soluble and transmembrane tumour necrosis factor. Blocking the action of soluble tumour necrosis factor, but not of soluble tumour necrosis factor and transmembrane tumour necrosis factor, protected mice against the clinical symptoms of experimental autoimmune encephalomyelitis. Therapeutic benefit was independent of changes in antigen-specific immune responses and focal inflammatory spinal cord lesions, but was associated with reduced overall central nervous system immunoreactivity, increased expression of neuroprotective molecules, and was dependent upon the activity of neuronal nuclear factor-κB, a downstream mediator of neuroprotective tumour necrosis factor/tumour necrosis factor receptor signalling, because mice lacking IκB kinase β in glutamatergic neurons were not protected by soluble tumour necrosis factor blockade. Furthermore, blocking the action of soluble tumour necrosis factor, but not of soluble tumour necrosis factor and transmembrane tumour necrosis factor, protected neurons in astrocyte–neuron co-cultures against glucose deprivation, an in vitro neurodegeneration model relevant for multiple sclerosis, and this was dependent upon contact between the two cell types. Our results show that soluble tumour necrosis factor promotes central nervous system inflammation, while transmembrane tumour necrosis factor is neuroprotective, and suggest that selective inhibition of soluble tumour necrosis factor may provide a new way forward for the treatment of multiple sclerosis and possibly other inflammatory central nervous system disorders.