Abstract: A fragile audio watermarking method for MPEG-2 AAC bitstreams was proposed. The algorithm of watermark embedding was carried out by modifying the LSBs (least significant bits) of quantized MDCT coefficient's magnitude that is greater than 15, and was improved by considering the watermark's distribution in frequency and the perceptual entropy of host audio. Experimental result shows the proposed watermarking scheme is not appreciable, and it is possible to insert high hide rate additive data effectively into encoded bitstreams. The embedding and retrieving process is easy and it is adaptive for real-time analysis applications.