In this paper, we discuss a feature reduction technique and their application to document clustering, showing that feature reduction improves efficiency as well as accuracy. We select the terms starting from the Goffman point, selecting an area of suitable transition making use for it of the Zipf law (our method is called NZIPF). Finally, we demonstrate experimentally that the transition zone that provides better results is taking 40 terms starting from the Goffman point for a cluster of documents with a genetic algorithm non-supervised. The experiments are carried out with the collection Reuters 21578 and the results are grouped by new genetic operators designed to find the affinity and similarity of the documents without having prior knowledge of other characteristics. © 2009 IADIS.