To solve Contourlet transform domain multiplicative watermark detection problem, the characters of high frequency directional subband coefficients in Contourlet transform were statistically modeled using generalized Gaussian distribution. Detection of the pseudo-random sequences with unknown parameters in generalized Gaussian distribution coefficients was used to formulate the framework of watermark detection. By using two-sided hypothesis testing theory, a blind detector applicable to Contourlet domain multiplicative watermarks was constructed. Experiments showed that the decision rule of the blind detector was valid for any embedding strength, and the detection threshold can also be determined adaptively.