Introduction: The availability of image derived input functions (IDIF) obviates the need for arterial blood sampling and thereby facilitates clinical use of quantitative PET studies. The aim of this study was to develop a method for deriving IDIFs using reconstruction-based partial volume correction (PVC) [1]. Methods: PET and arterial blood data from nine dynamic [11C]flumazenil scans, acquired using an ECAT EXACT HR+ scanner and an on-line blood sampler, were used to develop and evaluate the method. Scans were reconstructed using both standard (no PVC) ordered subset expectation maximization (OSEM, 2 iterations, 16 subsets) and a PVC-OSEM algorithm, which corrects for the spatial resolution of the scanner. Number of iterations and width of PVC kernel were varied. The following regions of interest (ROIs) Methods: were evaluated for defining cerebral arteries: (a) pixel value threshold, (b) variable number of 'hottest' pixels per plane, (c) region growing, (d) cluster analysis, and (e) MR-based ROI. ROIs were defined on a pseudo blood volume image, generated by summation of early frames (