The single diffusion tensor model is inadequate for the reconstruction of fiber pathways in brain regions with multiple fiber orientations. To overcome this limitation, constrained spherical deconvolution has been proposed. A high reliability of constrained spherical deconvolution is, however, a pre-requisite for its use in clinical applications. Reliability of reconstructed fiber pathways can be assessed in terms of architectural (addressing their spatial configuration) and microstructural (addressing diffusion-derived measures along the fibers) reproducibility. We assess the reliability for two clinically relevant fiber pathways: the corticospinal tract and arcuate fasciculus. The fiber pathways were reconstructed using constrained spherical deconvolution in 11 healthy subjects who were scanned on three occasions. Coefficients of variations of diffusion-derived measures were used to assess the microstructural reproducibility. Image correlation and fiber overlap were used to assess the architectural reproducibility. The mean correlation between sessions was 72% for both the corticospinal tract and arcuate fasciculus. The mean overlap between sessions was 63% for the corticospinal tract and 58% for the arcuate fasciculus. Coefficients of variations of diffusion-derived measures showed very low variation (all measures <3.1%). These results are comparable with reliability results based on the diffusion tensor model, which is commonly used in clinical settings. The reliability results found here are, therefore, promising to further investigate the use of constrained spherical deconvolution in clinical practice. Magn Reson Med 70:1544–1556, 2013. © 2013 Wiley Periodicals, Inc.