Thalamic nuclei can be distinguished by their characteristic fiber orientations, which influence the diffusion. Fiber orientations are relatively aligned within a nucleus due to the fact that the cerebrocortical striations within a nucleus all target the same region of cortex. The number of thalamic nuclei reported with histological methods varies with the method employed, although most cyto/myeloarchitec stains identify 14 major nuclei. We present a new approach for thalamic nuclei segmentation on High Angular Diffusion Resolution Images (HARDI), performed with a constrained k-means clustering. As described by John D.Carew, it is possible to classify HARDI data based on the shape of the diffusion, thanks to the complex information coming from them. Mette R. Wiegell  proposed a thalamic nuclei clustering with k- means on diffusion tensor images, using a combination of a voxel distance and a diffusion tensor distance. In the same way, we use the k-mean algorithm with a weighted sum of two distances to cluster the thalamic nuclei on HARDI data.