The maintenance phase of the software project can be very expensive for the developer team and harmful to the users because some flawed software modules. It can be avoided by detecting defects as early as possible. Software defect prediction will provide an opportunity for the developer team to test modules or files that have a high probability defect. Naïve Bayes has been used to predict software defects. However, Naive Bayes assumes all attributes are equally important and are not related each other while, in fact, this assumption is not true in many cases. Absolute value of correlation coefficient has been proposed as weighting method to overcome Naïve Bayes assumptions. In this study, Absolute Correlation Weighted Naïve Bayes have been proposed. The results of parametric test on experiment results show that the proposed method improve the performance of Naïve Bayes for classifying defect-prone on software defect prediction.