Abstract: With the rapid development of environmental policies in China, collating, inducing, analyzing and interpreting a large number of policies and regulations in a purely manual way has become more and more difficult. Therefore, it is of great significance to use computer technologies, such as text mining, to support intelligent environmental policy management and environmental policy analysis, including information extraction and text analysis. Accurate word segmentation, or tokenization, is the basis of all text mining functions. In order to improve the effect of policy text segmentation, the environmental policies published on official websites of China?s ecological and environmental departments of all levels were collected and taken as corpus. New word discovery algorithms and manual supplement and modification were adopted to develop the environmental management professional lexicon. The empirical results showed that with addition of the environmental lexicon, the accuracy of environmental policy segmentation could improve from 72.6% to 94.1%, and the misjudgment rate of policy automatic classification based on support vector machine could reduce by 22.7%. Besides, the results of word frequency statistics and keyword extraction after adding lexicon could also provide more comprehensive and more timely statistical information for environmental policy analysis.