We propose a new global registration method for estimating the cardiac displacement field in 2D sequences of ultrasound images of the heart. The basic idea is to select a reference frame (e.g., the first image of a cycle) and to map each image in the sequence to it using elastic deformation. What makes our method specific is the use of a semi-local parametric model of the deformation (spatio-temporal spline), and the reformulation of the registration task as a global spatio-temporal optimization problem. The scale of the spline model controls the smoothness of the displacement field. Our algorithm uses a multiresolution optimization strategy for higher speed and robustness. We validated the accuracy of our algorithm by applying it to a synthetic sequence; this one heart-cycle test sequence was generated by deforming a reference frame according to a realistic motion model and by adding random noise to it. Finally, we present results on real data from normal and pathological subjects to illustrate the clinical applicability of our method.