Spatial localization with the spectroscopic imaging technique is normally implemented with the Fourier-transform approach, yielding rectangular voxels, with potentially significant cross-voxel contamination. Multidimensional Fourier-series window (FSW) is an alternate approach that generates single voxels of predetermined shape, with minimal out-of-voxel contamination. The spatial location of the voxel is shifted by means of postacquisition processing. A two-dimensional circular voxel is introduced, which for many in vivo applications is a good match of the region of interest. Phantom images illustrate the spatial distribution of signal intensity within the circular FSW voxels. Phantom spectroscopic studies show excellent spatial localization, with no detectable out-of-voxel contamination. The circular FSW voxel approach is implemented in human and animal model studies, demonstrating the technique's utility. This arbitrary shape approach can be extended to three dimensions, defining, for example, cylinders, spheres, or ellipsoids.