The procedure for dividing the initial set of group expert assessments into subgroups containing homogeneous, consistent estimates has been proposed in this paper. The proposed approach allows to distinguish groups of experts, with a “close” opinion, to analyze them in order to develop a final (group) assessment that takes into account the opinions (arguments) of each expert. For dividing the expert group into subgroups with consistent opinions, the mathematical apparatus of the Dempster-Shafer theory was used. In contrast to existing approaches, this theory allows to take into account specific forms of un-factors, such as combination of uncertainty and vagueness arising in the process of interaction between expert judgments (evidence). Metrics of evidence theory characterizing the degree of proximity of expert assessments are taken as a measure of consistency. The Jousselme’s, Euclidean, Wang, Bhattacharyya’s and Tessem’s distances have been analyzed in this paper. Expert evidence is considered to belong to one group if the value of the specified metric (distance) for all evidence of this group does not exceed a specified threshold value. The farthest neighbor and average linkage methods have been used in paper to determine the order of grouping expert evidence in accordance with the degree of their similarity (dissimilarity). Numerical calculations of the proposed procedure of formation of consistent groups of expert evidences are provided. The results obtained allow to carry out a more profound analysis of the obtained expert information aimed at a synthesis an effective and substantiated group decisions.