A watermarking scheme for distinguishing different copies of the same multimedia document (fingerprinting) is investigated. Efficient transmission of natural data requires lossy compression, which might impair the embedded watermark. We investigate whether the quantization step in compression schemes can be replaced by dithered quantization to combine fingerprinting and compression. Dithered quantization offers the possibility of producing perceptually equivalent signals that are not exactly equal. The non-subtractive quantization error can be used as the watermark. We denote the proposed watermarking scheme as 'quantization watermarking.' Such a scheme is only practical for watermarking applications where the original signal is available to the detector. We analyze the influence of the dither signal on the perceptual quality of the watermarked document and the watermark detection robustness. Further, the cross-talk between the non- subtractive quantization errors for two different dither realizations is investigated. An analytical description for fine quantization and experimental results for coarse quantization show how the cross-talk depends on the characteristics of the dither signal. The derived properties of quantization watermarking are verified for combined JPEG compression and fingerprinting. The detection robustness for the proposed quantization error watermark is compared with that of an independent additive watermark.