Data Envelopment Analysis (DEA) requires that the data for all inputs and outputs are known exactly. When some outputs and inputs are known decision variables, such as interval data and ordinal data, the DEA models becomes a nonlinear programming problem and is called imprecise DEA (IDEA). When data assume to be interval, decision making units (DMUs) can divide to three classes as follows: (I) efficient in any cases, (II) efficient in maximal sense and inefficient in minimal sense, (III) always inefficient. In this paper, we discuss a new technique for assessing the sensitivity of efficiency and inefficiency classifications in DEA with interval data in above three cases. Also we find radius of stability for all DMUs. Available bank branch data in Iran was used to illustrate the applicability of this approach Mathematics Subject Classification: Operations Research, 90