Abstract: By making use of hidden Markov models（HMMs）, the recognition of uppercase and lowercase characters of bank bills is studied This includes forming the novel character segmentation algorithm and designing algorithms of HMM training and recognising In this HMM system some mistakenly handrwitten and homophonous characters used frequently also are regarded as different classes Meantime a method of smoothing HMM parameters is put forward in its training Experimental results show that this method is practicable and good application future for automatic bank bill recognition