A fault severity level identification method based on ordinal classification is proposed to identify the gear crack levels. The fault level identification is regarded as ordinal classification in which there are ordinal structures between different severity levels and some features have monotonic relationship with the severity levels. The feature evaluation and feature selection for fault severity level identification based on ordinal classification are discussed. Ranking mutual information is utilized to distinguish monotonic features and non-monotonic features of the original feature set, and then a feature selection algorithm is designed for ordinal classification when monotonic features are mixed with non-monotonic features. The experimental results demonstrate that the designed algorithm can select the features with high classification ability for classifying the crack fault severity. A fault severity recognition model is constructed using ordinal classification. The proposed feature selection algorithm can reduce the dimension of feature space, select the features with strong classification ability and improve the accuracy of fault severity level identification.