Dynamic functional connectivity (dFC) derived from resting-state functional magnetic resonance imaging sheds light onto moment-to-moment reconfigurations of large-scale functional brain networks. Due to computational limits, connectivity is typically computed using pre-defined atlases, a non-trivial choice that might influence results. Here, we leverage new computational methods to retrieve dFC at the voxel level in terms of dominant patterns of fluctuations, and demonstrate that this new representation is informative to derive meaningful brain parcellations, capturing both long-range interactions and fine-scale local organization. Specifically, voxelwise dFC dominant patterns were captured through eigenvector centrality followed by clustering across time/subjects to yield most representative dominant patterns (RDPs). Voxel-wise labeling according to positive/negative contributions to RDPs, led to 37 unique labels identifying strikingly symmetric dFC long-range patterns. These included 449 contiguous regions, defining a fine-scale parcellation consistent with known cortical/subcortical subdivisions. Our contribution provides an alternative to obtain a whole-brain parcellation that is for the first time driven by voxel-level dFC and bridges the gap between voxel-based approaches and graph theoretical analysis.