In cardiac dynamic PET studies, the input function can be obtained directly from the reconstructed images. Therefore, there is a need to convert the time accumulated count to the time-activity curve (TAC). Conventionally, this is done by dividing the total counts in a localized region on the reconstructed image obtained during each scan frame period by its frame duration. This conversion, however, can significantly bias the estimates of rate constants of a compartmental model describing the dynamics of a PET tracer. Three new methods are formulated in this study. These new methods either use the accumulated counts in the input function directly or convert the accumulated counts to the input function more accurately. Computer simulation results show, for C-11 acetate and F-18 fluoro-2-deoxy-D-glucose (FDG), that the three new methods proposed can improve significantly the parameter estimates over the ones obtained by the conventional method.