Fingerprint image is a special kind of gray image which consists with ridge and valley lines in black and white. The fingerprint image has different frequency component, low, medium and high frequency. The fingerprint image compression algorithm based on graded frequency is proposed. The low frequency sub-image, which contains the most energy, is encoded by lossless Differential Pulse Code Modulation (DPCM). The medium frequency sub-image, which contains less energy, is encoded by Embedded Zero-tree Wavelet (EZW). The high frequency sub-image, which contains the least energy, is encoded by Set Partitioning Embedded Block Coder (SPECK). The experimental result indicates, compared with the other image encoding method, the presented algorithm improves the compression ratio by about 21.4 percent and fidelity by about 6.25 percent in conditions of ensuring reconstruction quality.