The aim of this paper is to demonstrate that wavelet denoising processing is extremely attractive for efficient source separation of strong noisy mixtures. Systematic numerical simulations using source separation algorithms after wavelet denoising are used to provide quantitative evaluations of the efficiency of the method. The cases of correlated Gaussian and non-Gaussian noise are investigated, which open the way to various practical applications.