Data is grouped together based on similarity this technique is called clustering which very well known in datamining. For extracting use full data from cluster most of the people are using the algorithm K-Means. In K-Means approach selecting initial centroids is the problem & these centroids are selected randomly. Because of random centroids this algorithm re-iterate a many number of times. The K-Means algorithm Correctness depends much on the chosen central values. To enhance the performance of the K-Means one should not select the original centroids randomly these must be selected carefully. A new tactic to formulate the original centroids is proposed which improves the rapidity of clustering and cuts the computational complexity by reducing the number of iterations.