In dynamic cardiac PET FDG studies for measurement of myocardial metabolic rate of glucose (MMRGlc), the plasma FDG time-activity curve (input function) is commonly obtained from the left ventricular (LV) region on the PET images. The input function is contaminated by spillover of radioactivity from the surrounding myocardium and this could cause significant error in the estimated MMRGlc. In this study, we determined the effect of myocardial to blood pool spillover on MMRGlc and developed a method to correct for this spillover of activity. The method is based on a reformulation of the FDG model equation in terms of the spillover contaminated input function that includes both the myocardium to blood pool and blood pool to myocardium spillover fractions as variable parameters (Fmb and Fbm). The reformulated model equation can be used to fit the global myocardial tissue activity curve to estimate Fmb and thus yields a spillover corrected input function. The MMRGlc estimate with the corrected input function was within 95% of the true value (compared to 85% using the uncorrected input function) in a set of computer simulation studies. Dynamic PET FDG data were obtained in eight human studies and blood samples were obtained during the study. As compared to the results with the uncorrected input function, the estimates of k4 by the new method were reduced by 69% into a range consistent with in vitro results. The method is effective in correcting Fmb spillover and leads to more accurate estimates of MMRGlc. The method also allows larger regions of interest (up to 150 mm2) to be drawn over the LV in dynamic PET images, thereby reducing the noise level in the input function.