The continuous ultrasonic signal is a common acoustic phenomenon, and composite structures pose difficulties for traditional acoustic localization methods due to their anisotropy. To solve this problem, this paper proposes a minimum error indexing method based on frequency domain characteristic vectors. The method acquires signals through distributed sensors on a divided grid structure in advance, constitutes characteristic vectors, and generates an indexing matrix. When a new ultrasonic source occurs at an unknown location, the characteristic vector is formed by the same method and matched with the indexing matrix for ultrasonic source localization. This method is well adapted to complex structures like composite materials. The method was verified by 50 sets of localization experiments for chirp and white noise signals, The results show that the average error is 4.92 cm and 6.27 cm. It provides an idea for structure health monitor of composite materials in the future.