Abstract: In view of problem that the order of effective rank of matrix reconstruction is difficult to determine in singular value decomposition for noise reduction, a novel signal noise reduction method based on energy difference spectrum of singular value was proposed in order to improve signal to noise ratio of practical sample signal. According to the energy difference between useful signal and noise, the energy difference spectrum of singular value was constructed. The maximum peak point of energy difference spectrum curve was taken as valid order of the reconstructed signal, so as to realize separation between useful signal and noise, and the order of noise reduction of singular value was determined reasonably, integrity of the useful signal was retained and large signal-to-noise ratio was get, it's very important for signal feature extraction and analysis in next step. The results of simulation and example analysis show the effectiveness of the proposed method.