Early fault diagnosis can increase machinery availability and performance, reduce consequential damage, prolong machine life, and reduce spare parts inventories and breakdown maintenance. In this book, one intelligent fault diagnostic system is proposed based on feature extraction and selection techniques. Features are calculated from many domains: time domain, frequency domain, cepstrum domain and wavelet domain. In this way, the information of raw data is kept at best to meet different analysis methods in future. Principal component analysis and linear discriminant analysis, two feature extraction methods are introduced. Feature selection methods, individual feature evaluation and genetic algorithm, are compared. They are used to reduce feature dimensionality and improve system performance. The proposed system is applied to fault diagnosis of induction motors as a real application. The results show that the proposed system, combining feature extraction with feature selection, has fast training procedure, high classification rate and compact structure. It is suitable for motor condition monitoring and fault diagnosis.