Major　drawbacks　of　the　traditional　data　envelopment　analysis　(DEA)　method　include　selecting　optimal　weights　in　a　flexible　manner,　lacking　adequate　discrimination　power　for　efficient　decision-making　units,　and　considering　only　desirable　outputs.　By　introducing　the　concept　of　global　efficiency　optimization,　this　study　proposed　a　double　frontiers　DEA　approach　with　undesirable　outputs　to　generate　a　common　set　of　weights　for　evaluating　all　decision-making　units　from　both　the　optimistic　and　pessimistic　perspectives.　For　a　unique　optimal　solution,　compromise　models　for　individual　efficiency　optimization　were　developed　as　a　secondary　goal.　Finally,　as　an　illustration,　the　models　were　applied　to　evaluate　the　energy　efficiency　of　the　Chinese　regional　economy.　The　results　showed　that　the　proposed　approach　could　improve　discrimination　power　and　obtain　a　fair　result　in　a　case　where　both　desirable　and　undesirable　outputs　exist.