In view of the problem that useful characteristics of flood discharge structure vibration signal with low signal to noise ratio which extracted difficultly, a novel de-noising method, which combines wavelet threshold and empirical mode decomposition (EMD), was proposed. Firstly a part of white noises were filtered out with wavelet threshold method, which can reduce the endpoint effect of EMD; then decomposing the signal with EMD, obtained a series of intrinsic mode functions (IMF) which contains real physical meaning; finally reconstructed the IMF of characteristic information to achieve the de-noised signal through spectrum analysis. Constructing the simulation signal, and comparing the filtering effect of this method with digital filter, wavelet threshold and EMD, study shows that, it is a superior de-noising method, which can filter the vibration noise of flood discharge structure accurately and retain the characteristic information. It is used in Laxiwa arch dam hydro-elastic model, analyzing the dominate frequency of dam structure precisely, which can provide the basis for safe operation and on-line monitoring of the dam structure. This proposed method can effectively solve the problem of structure information extraction for large flood discharge structure under the background with strong noise.