This paper deals with the problem of optimizing two-stage structure decision making units (DMUs) where the activity and the performance of two-stage DMU in one period e ect on its efficiency in the next period. To evaluate such systems the e ect of activities in one period on ones in the next term must be considered. To do so, here a dynamic DEA approach presented to measure the performance of such network units. According to the results of proposed dynamic model the inefficiencies of DMUs improve considerably. Additionally, in models which measure efficiency score, undesirable outputs are mostly treated as inputs, which do not reect the true production process. This paper proposes an alternative method in dealing with bad outputs. Statistical analysis of sub-efficiencies, i.e. efficiency score of each stage, during all periods represents useful information about the total performance of the stage over all periods.