Block cipher algorithms are widely employed in the field of information security, and accurately identifying the encryption algorithm used in ciphertext has become a crucial task in block cipher analysis. Existing identification schemes typically use the same key for both encryption and testing during the training phase, generating ciphertext. However, in real-world systems, keys are frequently changed to meet different needs, and the characteristic patterns of ciphertext vary depending on the key used. This variability leads to poor generalization in models trained on datasets generated with a single key. To address this limitation, this paper proposes a block cipher algorithm identification scheme based on Fast Fourier Transform (FFT). By applying FFT technology, ciphertext is transformed from the time domain to the frequency domain, allowing the extraction of frequency-domain features. These features reveal unique ciphertext patterns generated by different encryption algorithms. In this study, four block cipher algorithms—AES, DES, IDEA, and ARIA—are selected, and multiple sets of different keys are used to encrypt plaintext, generating a diverse ciphertext dataset. Four mainstream machine learning methods, including XGBoost, are used for training and testing. The experimental results show that the binary classification accuracy of block cipher algorithms is approximately 96%, a 13% improvement over existing schemes based on randomness testing. The four-class classification accuracy is about 80%, showing a 10% improvement compared to existing CNN-based methods.