Naǐve-Bayes algorithm is a stable supervised learning method, and it is difficult to improve its predicting accuracy. This paper analyzes the properties of the error-correcting output codes generated by search-coding method at first, then presents a search coding based on vaǐve Bayes algorithm (SCNB), and describes the flow chart of SCNB in detail. Experimental results show that search-coding method is an efficient approach to improve the generalization for Bayesian classifiers.