Dynamic functional connectivity (dFC) based on resting-state functional magnetic resonance imaging (fMRI) explores the ongoing temporal configuration of brain networks. To reduce the large dimensionality of the data, conventional dFC analysis usually foresees an atlasing step, in which the brain is parcellated into specific regions of interest, and voxels' time-courses are spatially averaged within these regions before assessing connectivity. In this study, we addressed for the first time the exploration of dFC at the voxel level; i.e., without the use of any brain parcellation prior to the connectivity analysis. We used a sliding-window approach and extracted window-specific dominant patterns. To overcome the limitations due to the huge size of voxelwise connectivity matrices, we adopted the fast eigenvector centrality method with some adaptations to make it suitable for the dFC framework. After concatenation of the dominant patterns of all subjects, principal component analysis (PCA) was used to extract the eigenmaps; i.e., the most recurring voxelwise brain patterns characterizing resting-state. The obtained eigenmaps appeared consistent with previously observed resting-state eigenconnectivities, but with the substantial advantage of characterizing brain networks at the voxel level without the need of an atlas. The effect of the connection-wise temporal demeaning, usually performed in dFC analysis to remove the influence of static connectivity, was explored and does not seem to have an influence when voxelwise brain patterns are targeted.