This paper presents an adaptive additive watermarking model and analyzes corresponding detection performance. With the development of watermarking technology, more and more researchers focus on the adaptive algorithms, which use the diversity embedding strength according to the host image and watermark signal. These kinds of algorithms can achieve better unperceptiveness and robustness, and have become an important research branch so far. Therefore, modeling the adaptive watermarking algorithms is a very significant work. We construct a general model to describe the adaptive watermarking system in this paper. Meanwhile, the detection performance based on the normalized similarity measurement is analyzed in details. Two kinds of false possibility, i.e. the possibility of false positive and false negative, are deduced. And the relationships between them and decision threshold are given.