This paper presents a novel motion decoupling and registration method for 3D MR myocardial perfusion imaging. The technique uses tissue tagging for prospective through-plane motion correction of the left ventricle during 3D mulli-slice image acquisition. The remaining in-plane distortion of the heart due to respiration was corrected by using a multi-resolulion 2D free-form image registration method. Partial least square regression was adopted to recover the intrinsic relationship between respiration and cardiac deformation, both to speed up the registration process and to improve its internal consistency. Factor analysis is then applied to extract tracer characteristics of different regions of the myocardium. Both simulation data and in vivo images acquired from 8 normal subjects and 8 patients with coronary disease were used to validate the proposed lechniques.