The development of modal theory, modem testing and signal processing has jointly promoted a frontier structure detection technique using vibration and sound signals. The difficulty of this technique is to find a characteristic of the detected signals and an algorithm which can distinguish a great variety of different types and reflect the slightly difference between the defect and qualified structures of the same type. Now commonly used characteristics include natural frequency, damping ratio and vibration mode, and numerical comparison method cannot present the quantitative evaluating indexes. Moreover, the acquisition process is time consuming and expensive, thus, precluding industrial automation. This paper introduces a method of structure defect detection-- Frequency Response Similarity. The principle of Frequency Response Similarity algorithm is described in detail. To verify the validity of the algonthm, a test on a batch of ceramic tiles of various types and with various defectives was conduct. Frequency response signals from non-defective products of each type are used as the samples for training, then the testing signals are distinguished and assessed by their quantized similarity to the samples since a defect structures similarity to non-defective structure is significantly smaller than qualified ones. Test results show that the accuracy of this algorithm reaches 95%. Since it has advantages of less quantity of calculation and easily transplanting to the mini signal processor of DSP, the access to the applications on industrial detection line was conceived in this paper.