A cost function for adaptive filtering is proposed, based on the statistical reference criterion. The purpose of this criterion is to enforce a given probability density function at the output of a system. The only aprioristical knowledge utilized is that of the target PDF. This makes it particularly suitable for blind adaptation in the context of arrays in contrast to temporal or spatial reference techniques. Applications that are foreseen are: blind channel equalization, adaptive beamforming for single or multiple signals and source separation. The main goal is blind signal recovery in unfavourable wanted signal-to-inference power ratios