Abstract: Gives a blind signal and noise ratio(SNR) estimation method based on singular value decomposition of signal self-correlation matrix, and realizes the estimation of dimension of signal subspace by the principle of minimum descriptive length. Based on subspace decomposition ind, a method of SNR blind estimation based on power spectrum is proposed. We use power spectrum that is out of signal band to estimate the whole noise power. Then SNR value can be calculated out. This method does not need any prior information, and is suitable for all kind of non-constant envelope signals. The simulation results showed that when the SNR is lower than 3dB the method based on power spectrum works better than the method based on singular value decomposition.