Dosimetry for targeted radionuclide therapy has evolved in recent years and current research is focusing on the production of spatial dose distributions using 3D voxel-based approaches. In this work we propose the use of cluster analysis for 3D dosimetric applications as an automatic way to identify non-homogeneities by grouping voxels within a volume of interest (VOI) according to their functionality. We implemented k-means methodology with IDL and we applied it to 3D and to 4D simulated and patient data. The results showed that the proposed methodology could recover the non-homogenous regions, place automatically sub-VOIs over a VOI, and produce cumulated activity (CA) maps directly by fitting the centroids of the clusters or by a second step process using the clustered image as a map. An additional benefit of the proposed methodology could be its use to identify regions of excess noise originating especially from misregistration. An assessment of the full benefits and limitations of the proposed methodology remains to be done.