Abstract: A multi-region dynamic CGE model was established, which divides China into nine regions: Beijing-Tianjin-Hebei, Beijing-Tianjin-Hebei Rim Region, Central Region, Yangtze River Delta, Northeast, Northwest, Southwest, Pearl River Delta and Southern China. According to the severity of haze, China is divided into two groups: high-polluted areas and low-polluted areas. This paper set up three different simulation scenarios, and evaluated the policy impact from different perspectives: air quality, sulfur tax rate, emission reduction costs, total output of various industries and other dimensions. The simulation results are as follows.（1）In the baseline scenario, the air quality of nine regions will not meet the national first-level standard until 2030, and four of them will not even meet the second-level standard.（2）The sulfur tax rate stipulated by the current environmental protection law is far lower than the cost of emission reduction, which will not fundamentally reverse the haze problem.（3）The average cost and marginal cost of emission reduction are highly positively correlated with local economy’s dependence on energy exploitation. Coal-rich areas such as Northeast and Northwest have the highest cost of emission reduction, with the marginal cost of emission reduction reaching 500 000~700 000 Yuan/ton by 2030. Due to the low energy production and consumption, the cost of emission reduction in the Pearl River Delta and the southern region is the lowest among the nine regions, and the marginal cost of emission reduction in 2030 is only 300 000~450 000 Yuan/ton.（4）Sulphur tax and other environmental protection tax policies have led to a serious shrinkage of the coal industry. In the most severe simulation scenario（B03），total coal output will decline by more than 70% in 2030, compared with the benchmark scenario.（5）In the scenarios of implementing different policies in different regions（B01 and B02），high-polluting industries will change from tight-bound areas to wide-bound and unrestricted areas, that is to say, there exists the phenomenon of so-called “regulatory leakage”.