The capability of magnetic resonance imaging to measure and image molecular diffusion has provided a new source of image contrast and has proven extremely useful in assessing stroke-related damage and the fiber architecture of nerves and muscles. While 3D diffusion imaging would present significant advantages over 2D imaging, it has up to now been impeded by technical problems, such as long acquisition times and motion artifacts. The present application aims at solving these problems, to enable robust and practical 3D diffusion imaging.