There are many factors that affect the operation state in the wastewater treatment process. Generally, the probability of failure is much less than the probability of normal operation. Fault diagnosis of wastewater treatment is a high-dimensional unbalanced data classification. In this study, we propose a feature selection-based method to improve the classification performance of wastewater treatment fault diagnosis. Two filter-based feature selection methods and one wrapper-based feature selection method were used for experiments. Three classifiers of C4.5, Naive Bayes, and RBF-SVM were used to evaluate the proposed method. Experimental results show that the proposed method can significantly improve the overall classification accuracy and AUC value on the wastewater treatment fault diagnosis dataset.