The invention relates to a method for hidden danger and fault feature extraction for a train running system. The method comprises the following steps that (1) real-time operation data of a train are collected by a sensor installed on the train and speed signals in the running process of the train are obtained through a train-mounted network; (2) multiple sensor network data fusion processing is conducted on the real-time operation data and the speed signals, so that source data of hidden danger and fault feature extraction are obtained; (3) a feature extraction algorithm is conducted on the source data, so that feature quantities of different components of the running system are obtained; (4) a feature quantity value which is representative and capable of reflecting the operation state of the running system is selected from the feature quantities obtained in the step (3), the redundancy among the feature quantities is removed, and the final hidden danger and fault feature quantity of the running system is obtained. Compared with the prior art, the data extracted through the method is high in accurate and the running safety can be effectively improved.