The integrity and security of fingerprint images can be achieved using watermarking techniques. We introduce Maximum-Likelihood (ML) watermark detection method to detect an invisible watermark within discrete wavelet transform (DWT) coefficients of fingerprint images. The ML method, which is based on Bayes? decision theory and the Neyman-Pearson criterion, requires a probability distribution function (PDF), which must correctly model the statistical behavior of the DWT coefficients. The performance of the detector is tested by taking into account the different quality of fingerprint images. Both Generalized Gaussian (GG) and Laplacian models provide attractive results but with a slight superiority for the GG model.