Conventional kinetic models of brain glucose uptake and metabolism that visualize brain glucose as being in a single pool in equilibrium with plasma, are unable to account for some recently described experimental findings. These include microdialysis demonstrations of a brain extracellular fluid glucose concentration that is both low, and significantly affected by changes in neuronal activity; and observations of transient glucose export (transient negative whole-brain arteriovenous differences) in certain neuro-intensive care settings. A kinetic model that treats brain glucose as divided into more than one, kinetically distinct, compartment, implying the presence of a glucose "reservoir" behind the blood-brain barrier, and with plasma glucose initially entering a compartment other than the brain extracellular fluid, is more consistent with these experimental observations. Neuroanatomical considerations suggest that plasma glucose may initially exchange with an intracellular astrocytic glucose pool, rather than the brain extracellular fluid. Astrocyte glycogen, mobilized at times of increased neuronal activity, could form the reservoir whose presence is inferred from demonstrations of transient glucose export, but only if glycogenolytic products can be exported from astrocytes as glucose. This hypothesis is considered in the light of the frequently suggested concept of a "nutritional" role for perivascular astrocytes and invertebrate glia, taking up blood-borne glucose and passing on metabolic substrates to neurons. The implications of this model for 2-deoxyglucose-based methods for regional cerebral metabolic rate estimation are discussed. In general, errors due to the approximations inherent in the conventional three compartment kinetic model, may be expected to become less significant as metabolism is averaged over space and time. Thus the three-compartment model is probably acceptable for the description of metabolism at the relatively low spatial and temporal resolution of these techniques.