Diffusion spectrum imaging (DSI) has been demonstrated to resolve crossing axonal fibers by mapping the probability density function of water molecules diffusion at each voxel. However, the accuracy of DSI in defining individual fiber orientation and the validity of Fourier relation under finite gradient pulse widths are not assessed yet. We developed an ex vivo and an in vivo model to evaluate the error of DSI with gradient pulse widths being relatively short and long, respectively. The ex vivo model was a phantom comprising sheets of parallel capillaries filled with water. Sheets were stacked on each other with capillaries crossed at 45degrees or 90degrees. High-resolution T2-weighted images (T2W1) of the phantom served as a reference for the orientation of intersecting capillaries. In the in vivo model, manganese ions were infused into rats' optic tracts. The optic tracts were enhanced on T1-weighted images (T1WI) and served as a reference for the tract orientation. By comparing DSI with T2WI, the deviation angles between the primary orientation of diffusion spectrum and the 90degrees and 45degrees phantoms were 1.19degrees +/- 4.82degrees and -0.71degrees +/- 4.91degrees, respectively. By comparing DSI with the T1WI of rat optic tracts, the deviation angle between primary orientation of diffusion spectrum and optic tracts was -0.41degrees +/- 6.18degrees. In addition, two sequences of DSI using short and long gradient pulses were performed in a rat brain. The bias of the primary orientation between these two sequences was approximately 10degrees. In conclusion, DSI can resolve crossing fiber orientation accurately. The effect of finite gradient pulse widths on the primary orientation is not critical. (C) 2003 Elsevier Science (USA). All rights reserved.