In a multicenter study, high reproducibility among centers is essential. However, the construction of diffusion connectomes requires multiple complex steps. Errors may be easily hidden in the connectome matrix and possible deteriorate the intrinsic neural network analysis. The registration and parcellation based on the brain template are the final steps and are vital to the connectome matrix. A different anatomical and functional image-based atlas may produce variable reproducibility in diffusion connectome matrices. In this work, seven frequently used atlases were compared to test the reproducibility of different templates.