With the advancement in MRI, functional parcellation (FP) of brain structure(s) has become an important topic. However, the large number of voxels is a major obstacle. A-priori partitioning of the brain into several regions of interest (ROIs) is the main data-reduction strategy to simplify brain informatics. This study aims to examine the validity of ROI-based approach to FP by exploring the concordance of the relative distance structures between voxel-wise (raw data) and atlas-informed analyses. Structural and resting state functional MRI (rfMRI) scans of 26 right-handed healthy individuals were selected from the Rockland dataset. Four target regions were included in the analyses, that is, left and right thalamus and amygdala. For each voxel in the target region, four classes of correlation maps (sampling strategies) were constructed from the rfMRI: whole brain, cortex, 150 ROIs, and 70 ROIs (ROIs are informed by anatomical atlases). The relative distance metric between two different voxels was defined as the mean absolute difference of their associated correlation maps. Considering all the possible pairs of voxels in a target region, the relative distance structure was derived and stored in a matrix (distance map). For every target region, the distance maps were very similar across the four classes of sampling strategies, with the grand mean correlation coefficient reaching 0.95. The results confirm the validity of previous ROI-based analyses of rfMRI data in FP. The rationale and limitation are discussed and an analytic strategy of whole-brain FP is proposed.