The invention discloses an encrypted traffic identification method and device based on characteristic self-learning and is used for improving an identification rate of encrypted traffic in the prior art. The method comprises the steps of converting obtained current network traffic data packets into normalized numerical values or normalized gray values; storing the normalized numerical values corresponding to the current network traffic data packets in a text file; or generating a gray image according to the normalized gray values corresponding to the current network traffic data packets and storing the gray image in an image file; taking the text file or the image file as pre-trained test input of an encrypted traffic classifier; and identifying a traffic type of the current network traffic data packets through the encrypted traffic classifier.