The process of rotating machinery fault diagnosis is composed of signal acquisition, feature extraction and fault identification, among which feature extraction is the foundation of fault diagnosis and the key to obtaining accurate diagnosis results. To improve the sensitivity of the extracted features, a feature extraction method based on adaptive multiwavelets and synthesis distance evaluation index for rotating machinery fault diagnosis is proposed in this paper. For the ability of evaluating the sensitivity of feature parameters, the maximum value of synthesis distance evaluation index is taken as the optimizing object function, and the optimal multiwavelets are searched from the library of CL3 adaptive multiwavelets by genetic algorithm. Then they are used for extracting features from vibration signals of rotor. To prove the effectiveness of the proposed method, K-nearest neighbor classifier is used for analyzing the features extracted by the proposed feature extraction method, the synthesis distance evaluation index feature extraction method and the principal component analysis feature extraction method from vibration signals of the experimental rotating machinery under normal, unbalance, misalignment and rotor-to-stator rub conditions，respectively. The results show that the method proposed in this paper can improve the sensitivity of feature parameters and obtain higher fault recognition rate.