Introduction The tractography problem is concerned with the inference of neural connectivity from magnetic resonance diffusion images [1-3]. Canonically, neural connectivity is described by a connectivity matrix, expressing the degree of connectivity between any two locations in the brain. Here, we describe a method to derive cortical connectivity matrices from diffusion images. The method entails three steps: (i) diffusion spectrum imaging (DSI) [4] to resolve intravoxel orientational heterogeneity up to the cortical margin, (ii) solution of optimal paths between all pairs of points in a set of defined cortical surface points, and (iii) tabulation of the fitness of the optimal paths to construct a connectivity matrix. Given the interest of the present study in cortico-cortical connectivity we employed DSI to resolve the intravoxel fiber heterogeneity of the subcortical white matter. The DSI method, a relative of dynamic NMR microscopy [5], measures the microscopic spin displacement function in toto. The structure of the spin displacement function, which we refer to as the diffusion spectrum, can resolve complex intravoxel distributions of fiber orientation including white fiber intersections irresolvable by the tensor model. For example, at intersections of white matter fascicles the diffusion spectrum exhibits multiple discrete peaks with each peak directed towards a component fiber population. In addition to exhibiting complex, multi-modal structure in deep white matter intersections the diffusion spectrum exhibits complex, clearly non-mono-Gaussian structure at the subcortical margin [4]. To test the ability of DSI and the connectivity matrix algorithm to demonstrate relative connectivity within and between cortico-cortical networks, we mapped the connectivity matrix for a uniform series of points on the cortical surface spanning the visual and sensorimotor areas. Methods Image Acquisition Cardiac-gated, multi-slice, balanced-echo [6] diffusion images were acquired at 3T (Siemens, Allegra) on a normal ntervals, and 4×4×4mm voxel resolution. The balanced echo sequence consisted of 90-g-180-g-180-g-acq where the placement of the 180-pair was optimized to eliminate Eddy current distortions [6]. The diffusion gradient sampling scheme consisted of a keyhole Cartesian acquisition to include the set of points in q-space lying on a Cartesian grid !-1) within a sphere of radius qmax !-1, for a total of 500 sampling points [4] (Fig. 1). The above sampling scheme provided a diffusion spectrum resolution of (2qmax)-1 ! " # $%& '-1 ! The displacement spectra P(r) were obtained by discrete Fourier transform of the modulus of the echo signal as a function of q for each voxel.