Allocation of emission permits (AEP) is an important issue because of its significant effects on environmental governance and operations management. However, whether AEP results can constantly maintain systemic Pareto optimality still remains unclear and has not been investigated adequately in prior literature. We attempt to fill this gap by dividing the AEP process into a pre-stage observation process and a two-stage regulatory scheme as motivated by recent real-world examples. We apply the characterizations of each stage's state variables to build a conceptual AEP model based on the classical theory of data envelopment analysis (DEA). Considering different real-world scenarios, we extend this conceptual AEP model into three different AEP models: non-limited, uniform-limited, and heterogeneous-limited AEP models. The allocation schemes derived from each of these three models are proved theoretically to be systemic Pareto optimal in the corresponding scenarios. The advantages of our models over other AEP methods are real-world tractability, enforceability, and systemic Pareto optimality. We further conduct an empirical analysis on allocating SO2 emission permits among mainland China's major million-KW coal-fired power plants using the proposed models. Results of our empirical study show that the heterogeneous-limited AEP model exhibits higher performance over the non-limited and uniform limited AEP models. Thus, we suggest that the Chinese coal-fired power industry should employ the heterogeneous-limited AEP model in the practical allocation of SO2 emission permits. (C) 2017 Elsevier Ltd. All rights reserved.