To investigate whether a large SOE is associated with higher energysaving potential and CO2 abatement potential in China's power sector, we employ a Weighted Russell Directional Distance Function. This non-parametric non-radial DEA approach enables us to model bad output within an environmental production framework. The major strength of WRDDF is that it allows for non-proportionally adjustments of inputs and outputs. It thus makes it possible to investigate specific
factor's inefficiency levels. This innovative methodology is applied to a unique sample of Chinese coal-fired power enterprises located in the Zhejiang province in the years 2004 and 2008. Each power enterprise is assumed to produce electricity and CO2 with labor, capital and energy input. By comparing with the target reference on the piecewise frontier under the DEA technology, the maximum feasible energy-saving potential index (ESPI) and CO2 emission-abatement potential index (CAPI) for each power enterprise is estimated. Moreover, we perform statistical tests and find that large power enterprises are less efficient in 2004, but become more efficient in 2008 than small power enterprises in terms of energy utilization and CO2 emission. For the SOE, there is no statistical evidence on an inferior performance compared to non-SOE in 2004, but SOE exhibit higher efficiency than non-SOE in 2008. This dramatic change between 2004 and 2008 may result from the implementation of the “Top-1000 program” which set strict energy-saving targets for large power plants and SOE.