The low-frequency noise in the devices often appears to be a superposition of 1/f noise and white noise, which are induced by different physical mechanisms or microscopic defects. The separation of the two noises is necessary especially in the characterization of quality and reliability by means of noise measurement. The traditional separation method is usually used in frequency domain with limited frequency range, and the estimated value of exponent index γ for 1/f γ noise has considerable error when the ratio of 1/f noise and white noise is smaller than 20 dB. In this paper, based on wavelet denoising theory, a new method is proposed to separate 1/f γ noise from white noise in time domain and can be used to estimate γ values with much better accuracy even if the ratio of 1/f noise to white noise in low enough.