The invention discloses a power system transient state stability estimation inputted characteristic quantity selection method. The method comprises the following steps: operating data of the power system in different failure modes is extracted; an original input/output characteristic quantity set is constructed, and calculation is performed according to the extracted operating data to obtain initial input/output training sample data of a neural network; according to the initial input/output training sample data, the initial neural network structure is obtained through training; an initial neural network inputted characteristic quantity sensitivity matrix is calculated; an original inputted characteristic quantity importance index is calculated according to the sensitivity matrix; and a transient state stability estimation important inputted characteristic quantity is picked out according to the importance index. According to the method, not only the inputted characteristic quantity selection calculation process can be omitted to avide disadvantages of repeated original network training and improve the calculation efficiency; and but also the original characteristic quantity importance ranking can be used to reflect indexes related to the system operating state so as to simplify the neural network structure and improve the correctness of power system transient state stability estimation.