In this chapter we have reviewed several methods addressing the problem of source separation in various contexts, such as multi-channel, binaural, stereo or mono, with various degrees of a priori information or assumptions on the sources. To date, the results achieved are fairly good considering the complexity and ill-posedness of the problem but near-perfect separation is still beyond reach in most cases. In some applications, such as the addition of audio effects or re-spatialization of the sources, near-perfect separation is not a must as artifacts are less audible in the remixed sound. This field has undergone tremendous development in recent years, of which we could only present a partial view, and a considerable amount of work is still under progress.