Abstract: A new method is proposed to improve the performance of estimating the number of signal components in low Signal-to-Noise Ratio (SNR) environment, based on singular values of noise subspace and the principle of Constant False Alarm Rate (CFAR). Through researching the relationship between singular values of Hankel matrix and the energy of noise, the threshold of detecting singular values of noise subspace is got in a certain probability of false alarm by using the characteristic of Gaussian white noise envelope submitting to exponential distribution. Simulation results show that the method is effective in low SNR condition.