Abstract: Automatically recognizing predicate is one of the important research topics in shallow parsing.Based on the theory of “predicate is the head of Chinese sentence”，predicate’s context of Chinese sentences has been analyzed in detail，and the main factors that affects the presence of predicate have been discussed.Then，A probability model for recognizing predicate of Chinese sentences has been presented.The probabilities of quasi-predicates are estimated by maximum likelihood estimation.Discounted model is used to smooth parameters.Proposed method is tested with a small corpora，and the average accuracy of predicate recognition reaches 80.6% for adjective predicate and 83.2% for verb predicate，respectively.The experimental results show that this method is feasible and effective.