Abstract: Based on recurrent neural network and the nature of Chinese word and character, the input and output of place name recognition task were redefined and a label model of recurrent network was proposed for Chinese character level based on deep learning method. Compared with recurrent network based on word level, the model proposed based on Chinese character level in this paper, achieves significant improvement on precision, recalling and F value, the F value gets an improvement of 2.88%. When place names contain rare words, the model can improve the F value more to 26.41%.