Abstract: A novel audio watermarking scheme using wavelet decomposition and quantum neural networks (QNN) is proposed in this paper. Firstly, with the wavelet decomposition of the framed audio signal, the low frequency wavelet coefficients are mapped to the watermarking by QNN train. Then, the low frequency wavelet coefficients classified fuzzily are replaced with the ones classified accurately. In this way, the correct rate of the watermarking detection can be improved. Experimental results show that, by reasonable selection of threshold, the watermarking is robust against some different attacks effectively, such as noise adding, low-pass filtering, re-sampling and re-quantizing. Compared with BP neural networks, the correct rates of the QNN algorithm can be increased by an average of 1% without threshold.