A new method based on wavelet decomposition is presented for the restoration of turbulence-degraded images. For this method, two turbulence-degraded images are used as the inputs, for which the multi-scale decompositions are made using wavelet transform. The discrete values of the two turbulence PSFs in large scales can be estimated by mean of the Fourier frequency spectrum of the images in low frequency subbands. Removing blur is performed in the low frequency subbands of the images while reducing noise and preserving edges are made in the high frequency subbands. The experimental results show that the proposed method is highly effective for it not only greatly reduces the computational complexity and speeds up the restoration but also enhances the quality of restoration and the ability of resisting-noise well.