Fingerprint enhancement is an essential preprocessing step and it is crucial for the efficiency of the fingerprint recognition algorithm. An algorithm based on curvelet domain is proposed. The input image is decomposed into coarse and fine scales coefficients, and directional filters and soft threshold function are applied fully to enhance image and reduce noise. In curvelet domain, directional filters are used to enhance coarse scales coefficients. Meanwhile, soft threshold function is used to reduce noise of fine scales coefficients, which is much different from the conventional methods based on spatial domain and frequency domain. Experiments are carried out on FVC2004 databases. For fingerprint with bad quality, the results indicate that the proposed algorithm can better enhance fingerprint images and reduce noise than traditional methods, and need less time.