Argument Components Classification is an essential subtask in argumentation mining. Most machine learning methods improve the accuracy of the argument components classification by artificially constructing features. In this paper, the deep learning model of BILSTM-ATT-CNN-CRF is proposed. Firstly, the features can be learned automatically through the convolutional neural network, saving the cost of manual construction. Secondly, the combination of bilstm and attention network can better capture the contextual information between sentences. Finally, the Argument Components are classified by conditional random field. BiLSTM, BiRNN, and BiGRU network layers were used in experiments. It was found that bilstm had the best experimental performance. On the Persuasive Essays dataset published in Germany, this method is better than the baseline method.