The selection of the reference set in the frame of Disaggregation - Aggregation (D-A) methods (UTA*, UTA II, UTADIS), constitutes one of the most important steps of the process, while it influences the accuracy and reliability of the assessed preference model. The reference set ought to satisfy two conditions: a) the alternatives of the reference set should be familiar to the Decision Maker (DM) so as to express his/her preferences from a known situation; and b) the selected alternatives have to be representative of the total set, so that all the different points of view of the decision space to be taken into consideration. This paper presents a clustering technique that is embedded in the Multicriteria Decision Aid Systems MINORA and MIIDAS, which incorporates threshold of dissimilarity in order to support DMs or Decision Analysts (DAs) to select a representative reference set. This technique is compared with the most familiar multivariate data analysis clustering methods, such as k-means and hierarchical. Also, the technique is illustrated through a real world case study.