Performance of an application is a vital issue for user satisfaction. Performance bug refers to a specific kind of bugs that create lags and overheads in application execution. Often, it is difficult to localize and fix performance bugs due to insufficient developer knowledge regarding the characteristics of these bugs. Eliminating performance bugs manually from the application is time consuming and costly. This paper proposes a characterization and localization approach of performance bugs using Naive Bayes. It first detects the structures and behaviors responsible for the performance issues in source code. Then it learns the association of performance bugs with those coding characteristics from the training data set. Finally, it localizes potential performance bugs in source codes from the given bug report using Naive Bayes Classifier. Experimental result shows that our proposed approach can successfully predicts 75% of the performance bugs from various source codes.