A classification scheme assigns samples of the watermarked media to classes based on a classification criteria indicating a likely presence of a watermarked signal. Once classified, the scheme models a statistical distribution of the samples in each class. It then assigns a figure of merit to the samples in each class. A watermark detector and reader use the figure of merit to give greater weight to samples that are more likely to contain a watermark signal. Alternatively, the statistical distributions of the classes may be used to derive an estimate of a watermark signal in a pre-filtering stage of a watermark decoder. The watermark decoder then extracts a message from the estimate of the watermark signal.