Abstract: Tag tracking is a pre-step of the heart motion construction. Put forward in this paper is a new tag tracking method based statistical approach. The method uses the grid model as the basic structure. First, the method sees the grid nodes' positions as a Markov random field (MRF) model, and uses the EM algorithm to classify the nodes into two sorts by whether the node is in the ventricles. Then, the method designs different priori distribution and likelihood function for different sorts based on the function performed in the tracking progress. The method utilizes the iterated conditional modes (ICM) algorithm to do maximum a posteriori (MAP) estimation. The method is tested on several sequences of cardiac systole MRI. The result shows that the method can classify the grid nodes, so it can exactly track the SPAMM tag lines, and because of taking the grid model's Markov character into consideration, the grid can keep its topological shape during tracking progress.