Accurate estimation of cardiac motion from clinical medical images especially motion components parallel to edges has been found to be difficult using the intensity based optical flow algorithm without any prior information. This research aims to develop a new feature guided cardiac motion estimation method for gated myocardial perfusion (MP) PET images whose quality has greatly improved in recent years. A unique anatomical feature of human heart - the interventricular sulcus, which contains important motion information, is becoming visible and can potentially be extracted from MP PET images to guide the motion estimation process. In this work the motion of the extracted anatomical feature was used to create an initial estimate of the cardiac motion vector field (MVF) to be combined with the optical flow motion estimation algorithm to improve the estimation accuracy. To evaluate the proposed cardiac motion method, we simulated cardiac gated MP PET images from the 4-D XCAT phantom. The feature-guided initial MVFs were used in combination with the optical flow algorithm and applied to the simulated gated MP PET images to obtain the estimated cardiac MVF. The results were evaluated quantitatively by comparison with the cardiac MVF of the XCAT phantom which was regarded as the ground truth, and with those from non-feature based motion estimation method. The comparison indicates that our feature-guided method can achieve more accurate estimation of cardiac motion than non-feature-based method.