Cerebral blood flow is inversely related to in vitro whole blood viscosity, the major determinant of which is haematocrit. Haemodilution increases cerebral blood flow in polycythaemic patients and in subjects with high normal haematocrit. There is now increasing evidence that this relationship reflects a homeostatic and physiological regulation of oxygen-carrying capacity. A high normal haematocrit proves to be a weak risk factor for stroke whilst stroke risk is clearly related to the target haematocrit in patients treated for polycythaemia rubra vera. Whilst venesection remains accepted prophylactic treatment against stroke and other vaso-occlusive events in the latter case, no large scale trial has formally assessed the role of haematocrit reduction in patients with early manifestations of cerebrovascular disease like transient ischaemic attacks or in the early stages of multi-infarct dementia. There are theoretical reasons why a high haematocrit might have adverse effects on the cerebral circulation in the presence of vessel occlusion. Thus, flow and therefore oxygen delivery would become constrained by high viscosity (haematocrit) in the maximally dilated ischaemic vascular bed, and secondary thrombosis would be encouraged by low flow rates, and increased cell-cell interaction. These arguments have led to two large multicentre clinical trials of haemodilution in acute stroke victims. Neither has revealed any clinical benefit in the treated group. The reasons for the failure of the trials are discussed. It is envisaged that haemodilution, as well as retaining a clinical role in the prevention of stroke in patients with polycythaemia, may be used as an adjunct to other therapy for the immediate sequelae of cerebral ischaemia.