Cable sheath is one of the most common physical evidence at the scene of an explosion, as well as at other crime scenes such as kidnapping, theft, homicide and rape. In order to establish a method to identify cable sheath quickly and effectively, 40 cable sheath samples were tested by X-ray fluorescence spectrometer. In this paper, we examined and analyzed the cable sheath that may be left behind at various crime scenes such as explosions. The samples were preliminarily clustered by K-means clustering into 5 categories. On this basis, the multi-layer perceptron and Fisher discriminant analysis classification models are constructed. The results show that the accuracy of training set and test set of multi-layer perceptron classification model is 100%, and the accuracy of Fisher discriminant analysis classification model is 90% using the leave-one-out cross-validation method. It can be seen from the classification accuracy that the combination of X-ray fluorescence and machine learning methods can effectively classify cable sheaths quickly and accurately. Thus, this method can provide efficient traceability support for all kinds of crimes involving cable lines, such as explosion cases.