The present study deals with the analysis of double-indicator curves for blood-brain barrier studies. Two mathematical models which provide for the estimation of backflux of tracer from brain to blood in conjunction with heterogeneity of the cerebral capillary and large-vessel transit times were used for the analysis of D-glucose transport on the basis of cerebral venous outflow curves. The two models, non-mixed and well mixed, arise from differing assumptions regarding the effective region surrounding the capillary lumen. An approximate solution for the well-mixed model was developed to increase computation speed. Fourteen D-glucose outflow curves and their reference curves were obtained from nine patients and subsequently analyzed by the two models. Further, in five patients data were obtained under different physiological conditions: normal, decreased, and increased cerebral blood flow rates. The results support the appropriateness of the well-mixed model and heterogeneity of the cerebral capillary transit times. The median value for the average extraction was 0.18 and the median distribution space was 0.14. The latter value is similar to the brain extracellular space that has been estimated by other methods. The extraction values calculated from the peak of the venous outflow curves were significantly smaller than the whole-brain average extraction values estimated with the well-mixed model (0.157 vs 0.178, P less than 0.0005). In summary: (a) capillary heterogeneity is present in the human brain and changes with cerebral blood flow; (b) after crossing the blood-brain barrier, D-glucose distributes in the brain extracellular fluid; and (c) the extraction curve is significantly influenced by backflux.