The methods of group theory are applied to the problem of characterizing the diffusion measured in high angular resolution MR experiments. This leads to a natural representation of the local diffusion in terms of spherical harmonics. In this representation, it is shown that isotropic diffusion, anisotropic diffusion from a single fiber, and anisotropic diffusion from multiple fiber directions fall into distinct and separable channels. This decomposition can be determined for any voxel without any prior information by a spherical harmonic transform, and for special cases the magnitude and orientation of the local diffusion may be determined. Moreover, non-diffusion-related asymmetries produced by experimental artifacts fall into channels distinct from the fiber channels, thereby allowing their separation and a subsequent reduction in noise from the reconstructed fibers. In the case of a single fiber, the method reduces identically to the standard diffusion tensor method. The method is applied to normal volunteer brain data collected with a stimulated echo spiral high angular resolution diffusionweighted ( The sensitivity of the magnetic resonance (MR) signal to molecular diffusion provides the most sensitive noninvasive method for the measurement of local tissue diffusion characteristics (for a review, see Ref. 1). The basic effect from which diffusion information is derived is the signal diminution due to diffusive motions along the direction of an applied gradient field (2,3). The fact that diffusion can have a directional dependence was recognized early on (4), but there has been a great resurgence of interest in the application to diffusion-weighted imaging of human tissues, where inferences about tissue structure can be made from the directional dependence it imposes on the local diffusion. Anisotropic diffusion was first demonstrated in the brain by Moseley et al.,(5,6) and has been used to study a variety of other tissues (7). The determination of anisotropy requires a reconstruction of the local apparent diffusion, D app . If the diffusion process has spatially homogeneous Gaussian increments, the directional dependence can be completely characterized by the diffusion tensor D. This relates the signal loss along an applied gradient in an orthogonal Cartesian system defined by the imaging coordinate system to the diffusion along a direction in an arbitrarily rotated orthogonal system defined by the tissue (8). Reconstruction of local Gaussian diffusion can thus be posed as estimating the diffusion tensor (8), which, in principle, requires only six measurements plus an additional measurement for normalization (9). This technique is of particular interest in its application to the characterization of white matter tracts (10,11).However, it has long been recognized that the Gaussian model for diffusion can be inappropriate within the complex structure of human tissues (8,12). One way in which this model can fail is the presence of multiple fiber directions within a single imaging voxel. Because the single fib...