We investigated the quantitative accuracy of iterative and analytic reconstruction methods to determine myocardial blood flow (MBF) in PET studies in pigs using O-15 labeled water. Dynamic scans were acquired on the EXACT3D PET scanner on large white pigs after H215O injection (resting and dipyridamole-induced stress). Radioactive microspheres were injected during the H215O scans to provide a "gold standard" of MBF values. Data were reconstructed with 3DRP, FORE + attenuation-weighted OSEM, FORE-FBP and 3D-OSEM. Factor images were generated by cluster analysis and linear dimension reduction (myocardium and left and right cavities) and resliced to short axis images; 16 ROIs were defined in the left myocardium and 2 ROIs in the left and right cavities. ROIs were projected onto the dynamic images to extract time-activity-curves (TACs), which were then fitted to a single compartment model to estimate regional and global absolute MBF. Microsphere measurements were obtained in a similar way and 64 pairs of measurements were made. FORE-OSEM and 3D-RP overestimated the CFR values by 10% and 17%, respectively and 3D-OSEM and FORE-FBP underestimate CFR values by 3% and 17%, respectively. Regression analysis for 3D-OSEM and FORE-OSEM gave intercepts close to zero and slopes close to unity. Bland and Altman plots of differences between all reconstructions and Microsphere provided a mean difference and limits of agreements in favour of 3D-OSEM and FORE-OSEM (mean difference closer to zero and smaller SD). For absolute quantification of MBF using O-15 water, 3D-OSEM and FORE-OSEM proved to be at least as accurate as 3DRP and FORE-FBP compared with the Microsphere technique.