Abstract: Part of speech tagging is one of the crucial stages in the corpus annotation process.A simple recurrent neural network （SRNN） model is addressed in this paper. Based on the structure of a three layer feedforward network, feedback connections between hidden layer nodes and state nodes in the input layer are added to the SRNN. During the tagging process, the enhanced neural network model is capable of handling contextual information in the text.The training algorithm of SRNN is also discussed.After the network is well trained with annotated sentences, it can correctly tag 94% of the words in testing data.