Abstract: To solve the problems of high model dependence and low accuracy of sound source localization method based on geometric model in unstructured space, a localization method via sound-position fingerprint method was proposed. Position estimation was achieved by comparing the spatial characteristics of sound signal recorded at location point with the positioning database. The method included two stages. At the off-line sampling stage, sound signals of positioning reference nodes were obtained and the feature information was extracted to build the positioning database. While at the online stage, the real-time signal characteristics of pending sites were matched with the positioning database to accomplish the location task. Simulation results show that under the condition of sparse microphones and with large noise, the proposed localization method has rather small sound source position estimation bias, and the positioning performance can meet the demands of practical application.