Abstract: Traditional cross efficiency evaluation method lacks maneuverability due to that the preferential weight system is always not unique for some or all DMUs, and it lacks acceptability due to using the ultimate average cross efficiency scores to rank all DMUs. Current studies typically target deterministic distribution of index weights and ultimate cross efficiency scores assembling based on elimination of the assumption of average as two independent decision making problems and solve them respectively. In this paper, the evaluation model which combines self-evaluation and peer-evaluation is seen as a group decision making process, in which each DMU is treated as an "expert" and an opinion object simultaneously, and then an adaptive DMUs cross efficiency group evaluating algorithm is proposed. According to the close degree of evaluation results which are from each DMU and DMUs group, the algorithm gets "expert" weight for each DMU and deterministic preferential index weight systems for each DMU, which are used to evaluate themselves and other DMUs, in a single decision making process by iterative adjustments. The experimental verification and empirical research illustrate that algorithm proposed in this paper can efficiently converge, which can get objective and stable ultimate efficiency scores to rank all DMUs deterministically.

Key words: DEA(Data Envelopment Analysis), DMU(Decision Making Units), group evaluating, cross efficiency, adaptive