We address the problem of blind equalization of mixing channels, resulting from frequency selective fading and multipath in a multiuser CDMA system. We present a novel frequency-domain approach that employs second order spectral statistics of the observations and the users signatures. Unlike other methods, which are based on time-domain analysis, no particular assumptions are made about the support of the mixing channels, except that they have finite length. Since signatures correlations, rather that signatures themselves, are used, the proposed estimation is independent of users delays. Although performance does depend on the users-signatures, signatures orthogonality is not required.