After reviewing missing mechanism, methods classification, and several well known treatment methods for missing data handling, this paper proposes a new method NBI, Naive Bayesian Imputation. NBI models use the imputation attribute as class attribute to build NBC. In this way, the imputation problem is turned into classification problem. NBC is insensitive to missing data and can be improved by attribute selection strategy. Extensive experiments on datasets from UCI are conducted to assess the effectiveness of NBI.