One of the main challenges of performance evaluation in organizations and all systems is the irrationality and inaccuracy of the methods and criteria used. Traditional performance evaluation methods are mostly one-level, so they usually fail to provide sufficient feedback to identify inefficient units. Data envelopment analysis is a mathematical programming technique that compares the relative efficiency of several decision-making units based on observed inputs and outputs expressed by a variety of different scales. In practice, since many decision-making units are subdivided into smaller parts, with standard data envelopment analysis models that consider the organization as a whole, logical results are not obtained. Therefore, it would be better to use developed models like the two-stage DEA model to more accurately evaluate under investigation units in these conditions. Moreover, in cases that there are a large number of inputs and outputs, traditional DEA is not very efficient and it may consider a large number of units as efficient one. To deal with the problem, this study uses PROMETHEE method to rank criteria. After that, the efficiency evaluation problem is continued with most important inputs and outputs. Since the available information is usually incomplete and inaccurate, the problem is solved in the gray environment. The findings indicate a significant decrease in the number of identified efficient units which shows the improvement in discrimination power of DEA method. Additionally, the use of uncertain environment has led to more accurate estimates than previous definite models.