Diffusion Magnetic Resonance Imaging (DMRI) is an emerging technique permitting to visualize the neuronal architecture of brain white matter by measuring the diffusion of water molecules in tissues. A DMRI acquisition is composed of a collection of diffusion weighted images (DWIs) that characterize the diffusion property in several noncolinear directions. Resampling such acquisitions to obtain measures of diffusion in other directions is a problem that may arise when registering or comparing DWIs. In this paper, we present a comparison of several spherical interpolation schemes for DWIs. Numerical experiments are achieved on both synthetic and real data.