In order to increase the signal to noise ratio of frequency response f unction (FRF) in field test, an effective FRF noise reduction method based on si ngular value decomposition (SVD) was presented. Firstly, the phase space reconst ruction of the impulse response function was established, and then the trajector y attractor matrix was decomposed by using SVD. The optimal denoising order wa s determined according to the increment of singular entropy. A numerical simulati on with closely spaced modes was employed using the GARTEUR plane model. The res ults show that the accuracy of estimated modal parameters is improved obviously from the denoised FRF, especially for the damping ratio accuracy.