Considering that the legal amount, courtesy amount and E13B amount code on the Chinese bank checks can be recognized independently, this paper presents an information fusion based check recognition system called IFCRS. An adaptive template matching algorithm is presented to locate recognition regions for check layout analysis. We use hidden Markov model to segment and recognize legal amount, and adopt artificial neural network to recognize courtesy amount and E13B code, respectively. D-S evidence theory is introduced to fuse above three recognition results for better recognition performance. Experimental results demonstrate that the system can robustly recognize checks and the information fusion based algorithm improves the recognition rate by 5~10 percent.