Suboptimal temporal sampling of left ventricular (LV) blood pool and tissue time-activity curves (TACs) may introduce bias and increased variability in estimates of myocardial blood flow (MBF) and flow reserve (MFR) from dynamic PET myocardial perfusion images. We aimed to optimize temporal sampling for estimation of MBF and MFR. Twenty-four normal volunteers and 32 patients underwent dynamic stress/rest rubidium-82 chloride (82Rb) PET imaging. Fine temporal sampling was used to estimate the full width at half maximum (FWHM) of the LV blood pool TAC. Fourier analysis was used to determine the longest sampling interval, T S, as a function of FWHM, which preserved the information content of the blood phase. Dynamic datasets were reconstructed with frame durations varying from 2 to 20 seconds over the first 2 minutes for the blood phase and 30 to 120 seconds for the tissue phase. The LV blood pool and tissue TACs were sampled using regions of interest (ROI) and fit to a compartment model for quantification of MBF and MFR. The effects of temporal sampling on MBF and MFR were evaluated using clinical data and simulations. T S increased linearly with input function FWHM (R = 0.93). Increasing the blood phase frame duration from 5 to 15 seconds resulted in MBF and MFR biases of 6-12% and increased variability of 14-24%. Frame durations <5 seconds had biases of less than 5% for both MBF and MFR values. Increasing the tissue phase frame durations from 30 to 120 seconds resulted in <5% biases. A two-phase framing of dynamic 82Rb PET images with frame durations of 5 seconds (blood phase) and 120 seconds (tissue phase) optimally samples the blood pool TAC for modern 3D PET systems.