The process of protein modeling usually involves the production of a variety of structures requiring efficient tools for structure model comparison attempting to choose the best three-dimensional (3D) structure. This paper introduces an alternative method for clustering 3D protein models that, instead of using attributes related to structural alignment to group the data, use quality-attributes of those models to represent and cluster them. This method stands out by removing the need to define a priori a base model for structural alignments. Even so, it is possible to present the most representative structure in each cluster, which is useful for docking or molecular dynamics studies. All the results were statistically analyzed and compared with decisions made by professionals to validate the proposed algorithm. The experiment simulated a usual protein comparative modeling process for different CATH classifications. The calculated variance levels after the dimensional reduction validate the workflow for different protein chain sizes. All the molecular descriptors for the input files are calculated by MHOLline 2.0, an online scientific workflow for studies on bioinformatics and computational biology, available for free on www.mholline2.lncc.br, or hand made using specific programs (e.g., MODELLER, PROCHECK) and adjusting the data to the template specified in this document. The Quality-Model Clustering Tool (QMC) and the data set used in this work are available for download on the git repository github.com/ruanmedina/Quality-Model-Clustering.