Abstract: A new algorithm for embedding a watermark into a digital audio signal is proposed in this paper. The watermark used in the algorithm is not the conventional pattern, such as pseudo-random sequence, but a visually recognizable binary image. First, convert the two dimensional image to a one-dimensional binary sequence. Then divide the digital audio signal into segments and perform discrete cosine transform (DCT) on them. At last, the watermark is embedded into the audio signal by modifying its middle-frequency coefficients in DCT domain. The technique of pseudo- random permutation is used during watermark embedding process. The experimental results show that the algorithm is robust to many operations to digital audio signal, such as filter, compression and so on.