Abstract: Discretization is an effective technique to deal with continuous attributes for machine learning and data mining.Reasonability of a discretization process determines the accuracy of expression and extraction for information.Dealing with the discretization of real value attributes，Chi2 algorithm can get a good result of the conflict-free data but do not well in inconsistency and incomplete data.This paper makes full use of the Bayseian model which allows for the wrong classification in nature and improved the Chi2 algorithm.The improved algorithm is not only more suitable for inconsistency and incomplete data，but also make the interval merging more reasonable.The experimental results have proven the validity of the new algorithm.