In this paper, we present a new thalamus segmentation method for MRI images based on a new type of deformable model-Lagrangian Surface Flow. Given a MRI image, the user can interactively initialize a seed model within region of interest. The model will then start to grow according to both boundary and region information based on the principle of variational analysis. The deformation will stop when an equilibrium state is achieved. Our experiments demonstrate that the new method is robust to image noise and inhomogeneity and will not get stuck into local minima or leak from spurious edge gaps.