Watermarking is a technique by which additional information can be embedded into the content of a signal itself, rather than the header, without any distortion that is noticeable to humans. It can also be regarded as a method to provide a certain rule to the content, which rule is difficult to ascertain by other persons. With ordinary audio watermarking techniques, such as amplitude modulation , phase modulation   and artificial echoes unnoticeable to humans produced by a specified impulse response    are used as the rule. This rule is often something that can be detected by means of a method of data mining. Hence, in this paper, Association Analysis , which is one of the typical analysis methods of data mining, is applied for this purpose. Items that are used in Association Analysis are denned using the MDCT or FFT coefficients of an audio signal. To embed a watermark, these coefficients are manipulated so that the result of Association Analysis of the watermarked signal has a singular pattern. Computer simulations showed that it is possible to embed a watermark by this proposed method without any noise that is noticeable to humans.