O2 metabolism is a crucial biomarker of brain physiology. We aim at measuring the repeatability of estimates of O2 metabolism obtained with a dual calibrated fMRI experiment and a newly proposed parameter estimation approach based on a forward model. The analysis is carried out on two datasets from volunteers scanned at rest. The performances are shown to depend on the physiological parameter considered, resolution (bulk or voxel-wise level) and decreasing from within- to between-sessions repeatability. Overall the repeatability is demonstrated to be comparable with PET and a previous calibrated fMRI method, but supplying a more complete mapping of brain O2 metabolism.