Abstract: Aimed at the problems of insufficient ciphertext feature extraction and low recognition accuracy in existing cryptographic algorithm recognition, a FBR ciphertext feature extraction method is proposed. This method combined the three methods of frequency, block frequency and run in the randomness test to define the statistical value of the number of symbols in the ciphertext, the statistical value of the number of runs, and the statistical value of the number of times within the block. FBR features were constructed based on three statistical values. The experiment used support vector machines to perform ciphertext two-classification and multiclassification experiments on the three mixed data sets. The experimental results show that the FBR ciphertext features extracted by this method are compared with the ciphertext features that have performed well, and the average recognition accuracy is improved, which fully proves the effectiveness of the proposed method.