This paper proposes a Chinese medical question answering system based on CNN s with self-attention embedded model. The key idea of the proposed framework is first understanding whole sentences via LSTM, obtaining the feature map by the convolution layer of CNN, then further-comprehending of question and answer by self-attention and finally pass through to polling layer of CNN to enhance accuracy. In addition, due to the word segmentation issue in the Chinese especially in medical field, character embedding is applied to enhance the accuracy. The experimental results show that the proposed framework improves the accuracy around 80%.