To facilitate automatic segmentation, we adopt SVM (Support Vector Machine) to localize the left ventricle, and the segmentation is then carried out with narrow band level set. The method of generating the narrow band is improved such that the time used is reduced. Based on the imaging characteristics of the tagged left ventricle MR images, BPV (block-pixel variation) and intensity comparability are introduced to improve the speed term of level set and to increase the precision of segmentation. Our method can perform the segmentation of the tagged left ventricle MR images accurately and automatically.