In this paper, a new algorithm for blind inversion of Wiener systems is presented. The algorithm is based on minimization of mutual information of the output samples. This minimization is done through a Minimization-Projection (MP) approach, using a nonparametric “gradient” of mutual information.