With the increasing use of the Internet and electronic documents, automatic text categorization becomes imperative. Many classification methods have been applied to text categorization. The k-nearest neighbors (k-NN) is known to be one of the best state of the art classifiers when used for text categorization. However, k-NN suffers from limitations such as high computation, low tolerance to noise, and its dependency to the parameter k and distance function. In this paper, we first survey some improvements algorithms proposed in the literature to face those shortcomings. And second, we discuss an approach to improve k-NN efficiency without degrading the performance of classification. Experimental results on the 20 Newsgroup and Reuters corpora show that the proposed approach increases the performance of k-NN and reduces the time classification.