Document clustering is the task of grouping a set of documents into clusters so that the documents in the same cluster are similar to each other than to those in other clusters. One of the applications of document clustering is in web search engine retrieval system to help the users find relevant information quicker, and allow them to focus their search in the appropriate direction. K-means is a commonly used algorithm for document clustering, but it has some disadvantages. The main limitations of K-means are: 1) The number of clusters K has to be given as input and 2) Based on the initializations it converges to different local minima. 3) It is slow and cannot be used for large number of data points.4) It cannot handle empty clusters. In this paper, we have developed a novel algorithm to eliminate all these basic drawbacks of K-means.