The order of effective rank and the row number of reconstruction matrix both are difficult to determine for noise reduction based on singular value de composition. In order to increase the signal to noise ratio of test signals, a new effective method is presented to solve this problem. According to the relati onship between signal frequency and the distribution of singular value, the order of effective rank is determined by the number of dominating frequency in the fast flourier translation result of the signal and the optimal row number of rec onstruction matrix is determined by the signal to noise ratio (SNR) and the mea n square error (MSE) of the denoised signal. A numerical simulation using differ ent signals with different frequency components is employed. The results show that the order of the effective rank is twice as the number of dominating freq uency and it does not change with the variation of the row number. The optimal r o w number can be defined as half the length of the signal in engineering applic ation. This method both has good noise reduction accuracy and stability with whi te noise and colored noise when the dominating frequency can be distinguished in the power spectrum.