Diffusion MRI is an in vivo and non invasive imaging technology that uses water diffusion as a proxy to probe architecture of biological tissues. Diffusion MRI has been widely used in white matter fiber tracts reconstruction in brain as well as many clinical applications including Alzheimer's disease. In this talk, We discuss various statistical models for diffusion MRI data. These models are used to estimate local neuronal fiber organizations, which are then used as inputs in tracking algorithms to reconstruct white matter fiber tracts. We will discuss their capability in resolving crossing fibers -- a major challenge in diffusion MRI. We will also consider spatial smoothing schemes to leverage information from neighboring brain voxels. These methods are illustrated through both synthetic experiments and diffusion MRI data from large brain imaging consortium.