Abstract Cerebral blood flow (CBF) obtained by dynamic susceptibility contrast MRI (DSC-MRI) often provides an abnormal contrast between normal and ischemic areas, as compared with those from positron emission tomography (PET) or SPECT. A linear relationship is normally assumed in the transverse relaxation rate change in relation to the regional concentration of the contrast agent. This study was intended to evaluate this assumption. The impact of an alternative, non-linear model based on a realistic capillary structure was also evaluated in the assessment of CBV and CBF by use of DSC-MRI. DSC-MRI and 15O-PET were carried out on 13 patients with chronic cerebrovascular disease. Regions-of-interest (ROI) were selected both in the ischemic and unaffected hemispheres, and the left-to-right (L/R) ratios of CBF and CBV were compared between PET and DSC-MRI. When assuming the linear relationship, the contrast of CBF by DSC-MRI was lower than that by PET (CBF L/R,MRI=0.60CBF L/R,PET+0.40, r=0.63, p<0.0216), and the contrast of CBV by DSC-MRI greater than that by PET (CBV L/R,MRI=1.34CBV L/R,PET−0.33, r=0.84, p=0.0003). Disagreement of CBF was particularly apparent in the area of enhanced CBV. The employment of the non-linear model improved the agreement between DSC-MRI and PET for both CBF and CBV (CBF L/R,MRI=0.89CBF L/R,PET+0.08, r=0.93, p<0.0001; CBV L/R,MRI=1.03CBV L/R,PET−0.04, r=0.79, p=0.0014). These results suggest that the linear model has limited accuracy in the assessment of CBF and CBV in patients with chronic cerebrovascular disease, and that the non-linear correction is essential in order to provide accurate quantitation with DSC-MRI.