This paper proposes to use DEA models with undesirable outputs to construct the Malmquist index that can be use to investigate the dynamic changes of CO2 emission performance. With the index, the authors have measured the CO2 emission performance of 28 provinces and autonomous regions in China from 1996 to 2007; with the convergence theory and panel data regression model, the authors analyze the regional differences and the influencing factors. It is found that the performance of CO2 emissions in China has been continuously improved mainly due to the technological progress, and the average improvement rate is 3.25%, with a cumulative improvement rate of 40.86%. In addition, the CO2 emission performance varies across four regions. As a whole, the performance score of eastern China is the highest. The northeastern and central China has relatively lower performance scores, and the western China is relatively backward. The regional differences are decreasing, and the performance of CO2 emissions is convergent. The influence of some factors on the performance of CO2 emissions is significant, such as the level of economic development, the level of industrial structure, energy intensity, and ownership structure. The influence of some factors, such as opening-up to the outside world, on the performance of CO2 emissions is not significant.