The invention discloses a method for extracting a sensor network skeleton, which comprises the steps of: 1, finding angular points on the boundary, and dividing the boundary into limited boundary branches; 2, identifying interconnected skeleton nodes; 3, searching two skeleton nodes with largest distance in the interconnected skeleton nodes, and connecting the two skeleton nodes to generate a skeleton chord; 4, connecting adjacent skeleton chords, and then connecting the angular points with nearest skeleton chords to generate a rough skeleton drawing; and 5, adopting a pruning method to optimize the rough skeleton drawing to obtain a final skeleton. The method utilizes different boundary branches to determine the skeleton nodes; and compared with the conventional algorithm, the method cannot be affected by boundary perturbation, so a more approximate network topology structure can be obtained, thereby better reconstructing a network.