The percussion-based method refers to a detection technique that utilizes a hammer to percuss test structures to produce vibration and sound, and diagnoses any abnormalities in structures based on sound features. Percussion-based bolt loosening detection has been one of the hot research topics in recent years. Most existing studies focus on the issue of single bolt loosening, while multi-bolt connections are commonly used in engineering applications. Therefore, current methods can not meet requirements for multi-bolt loosening detection in engineering. This paper proposes a fast percussion-based multi-bolt loosening detection method, using a single-row multi-bolted steel beam-column joint as the research object. Firstly, characteristic vectors of percussion sound of each bolt within the joint are extracted by short-time Fourier transform and binarization processing. Then, the Euclidean distances between characteristic vectors of each bolt under the loosening cases and under the reference case are calculated to quantify loosening characteristics. Normalized loosening indexes are subsequently constructed by combining loosening characteristic values of all bolts to measure loosening degree of the joint. Finally, the loosened bolts are identified by comparing the difference between loosening characteristic values and baseline value of each bolt. Experimental results show that the proposed method can rapidly and accurately identify loosened bolts in the joint, and further enhance the potential of percussion detection method in engineering application.