While H2(15)O positron emission tomography (PET) is still the gold standard in the quantitative assessment of cerebral perfusion (rCBF), its technical challenge, limited availability, and radiation exposure are disadvantages of the method. Recent work demonstrated the feasibility of magnetic resonance (MR) for quantitative cerebral perfusion imaging. There remain open questions, however, especially regarding reproducibility. The main purpose of this study was to assess the accuracy and reproducibility of MR-derived flow values to those derived from H2(15)O PET. Positron emission tomography and MR perfusion imaging was performed in 20 healthy male volunteers, who were chronic smokers, on day 1 and day 3 of a 4-day hospitalization. Subjects were randomly assigned to one of two groups, each with 10 subjects. One group was allowed to smoke as usual during the hospitalization, while the other group stopped smoking from day 2. Positron emission tomography and MR images were coregistered and rCBF was determined in two regions of interest, defined over gray matter (gm) and white matter (wm), yielding rCBF(PET)gm, rCBF(MR)gm, rCBF(PET)wm, and rCBF(MR)wm. Bland-Altman analysis was used to investigate reproducibility by assessing the difference rCBFday3 - rCBFday1 in eight continual-smoker volunteers. The analysis showed a good reproducibility for PET, but not for MR. Mean +/- SD of the difference rCBFday3 - rCBFday1 in gray matter was 6.35 +/- 21.06 and 0.49 +/- 5.27 mL x min(-1) x 100 g(-1) for MR and PET, respectively; the corresponding values in white matter were 2.60 +/- 15.64 and -1.14 +/- 4.16 mL x min(-1) x 100 g(-1). The Bland-Altman analysis was also used to assess MRI and PET agreement comparing rCBF measured on day 1. The analysis demonstrated a reasonably good agreement of MR and PET in white matter (rCBF(PET)wm - rCBF(MR)wm; -0.09 +/- 7.23 mL x min(-1) x 100 g(-1)), while in gray matter a reasonable agreement was only achieved after removing vascular artifacts in the MR perfusion maps (rCBF(PET)gm - rCBF(MR)gm; -11.73 +/- 14.52 mL x min(-1) x 100 g(-1)). In line with prior work, these results demonstrate that reproducibility was overall considerably better for PET than for MR. Until reproducibility is improved and vascular artifacts are efficiently removed, MR is not suitable for reliable quantitative perfusion measurements.