Fusion of wide-area traveling wave (TW) information in the power network can improve the accuracy and reliability of fault location. However, there is no available multi-source fusion location method which can adaptively determine the importance of each TW measurement point at different fault locations according to network topology. Hence, a novel fault location method based on wide-area TW and graph attention network (GAT) is proposed in this paper. Firstly, all TW measuring points of the whole network and the overhead lines among them are taken as the nodes and edges to construct the graph data of wire-area TW, while the variational mode decomposition (VMD) theory is employed to obtain the nodes′ features. Then, GAT mines the wide-area TW characteristics according to the topology correlation of the power network, identifies the fault line, and outputs the adaptive weight that represents the importance of the measured points. Furthermore, the adaptive weight is used to fuse multi-source TW information to calculate the precise fault location. The results show that this approach can locate multiple types of line faults, and the location errors at many typical fault points are within 100 m. Compared with the traditional methods, this approach has more obvious advantages in case of fewer measuring points, and the location accuracy under different fault points is improved by 20~ 400 m.