In order to reduce false-positive error probability and false-negative error probability for bills′ watermarks detection, an algorithm based on prior information and multi-template matching is proposed. By means of stepped-matching idea, it selects initial watermark template and uses single-template matching algorithm to find the approximate position of initial watermark template firstly. Secondly, it uses prior information to calculate watermarks′ scattering region. Thirdly, it uses multi-template matching algorithm to detect watermarks in the scattering region. Finally, it uses prior information to check the watermarks′ position that been detected, and to estimate the watermarks′ position that not been detected. The experiments show that the new algorithm can reduce the false-positive error probability and false-negative error probability, and it is effective to detect bills′ watermarks.