K-means is a widely used partitional clustering method. While there are considerable research efforts to characterize the key features of K-means clustering, further investigation is needed to reveal whether the optimal number of clusters can be found on the run based on the cluster quality measure. This paper presents a modified K-means algorithm with the intension of improving cluster quality and to fix the optimal number of cluster. The K-means algorithm takes number of clusters (K) as input from the user. But in the practical scenario, it is very difficult to fix the number of clusters in advance.