With the growing problem of data overabundance, many techniques for knowledge discovery have been formulated. Document clustering is one such method. This method aims to classify documents into meaningful categories that allow the user to quickly browse to pertinent information. Current techniques employ a word by word comparison of each document to be clustered. Certain “stop words” are not included in this process; determiners, conjunctions, prepositions, and pronouns do not lend any additional information to a sentence. This can be very computationally very expensive. We propose a new method of feature selection for document clustering which allows us to perform fewer comparisons. Proprietary issues require me to be no more descriptive than that at this time. Preliminary results suggest a 9% increase in accuracy over the established technique, as well as an 82% reduction in comparisons. Because only the most significant words are being used for clustering, documents are more likely to be clustered correctly. This drastic reduction in comparisons also means that this technique is much less computationally complex.