Recently, fingerprint has been received considerable attentions as a user authentication technique using biometrics. Fingerprint retrieval, which retrieves a candidate list of fingerprints having similar features with the given fingerprint from a fingerprint database, is one of interesting real-time applications. However, using the straightforward approach, it takes a long response time to scan the entire database and to compare the query against each reference on a large-scale database. Even when implemented on a hardware, it is hard to satisfy the real-time requirement. In this paper, to reduce the response time, we employ the geometric hashing technique which has been widely used for model-based object recognition. Using this algorithm, the number of fingerprint minutiae can be restricted. It limits the hardware size used for implementation to fit into FPGA-chips. Furthermore, we develop a scalable implementation for parallel geometric hashing on an FPGA-based PCI platform.