In the case of metrics-based software defect prediction, an intelligent selection of metrics is one of the key factors that affect the model performance. To solve the problem that only the correlation between software metrics is considered and the issue of redundance is tend to be ignored in the current studies, a new algorithm which combines ReliefF feature selection algorithm and correlation analysis is proposed. An experiment via 3 different classifiers over classic data sets from PROMISE repository is carried out, which is compared to the other two well-known feature selection algorithms. The ANOVA (Analysis of Variance) analysis shows that, a new method called ReliefF-LC (a fusion algorithm based on ReliefF and linear correlation analysis) feature selection algorithm can improve defect prediction performance.