A new use of neural network (NN) computation is proposed for DEA performance analysis. the proposed approach produces a smooth production frontier that is considerably different from the conventional piece-wise linear approximation. The smooth production frontier provides DEA users with a new evaluation basis for ranking analysis. The ranking analysis can be explored under variable returns to scale (RTS). The methodological feature is unique and important, because the traditional DEA ranking analysis usually assumes constant RTS on a production possibility set. The DEA ranking analysis often produces in infeasible solution when it is measured under the variable RTS technology. The proposed approach avoids such a difficulty. Two illustrative examples are examined in a combined use of DEA/NN.