Clustering is a process used to divide data into a number of groups. All data points have some mathematical parameter according to which grouping can be done. For instance, if we have a number of points on a twodimensional grid, the x and y coordinates of the points are the parameters according to which clustering is done. If the k-means algorithm is run with k=3, the data points will be split into 3 groups such that the sum of the variance for each group is minimized. The problem here, of course, is the choice of the parameter k. We may get a much better modeling of the data if we split the data points into 2 or 4 groups. Determining the ‘best’ value of k is a broad problem – there is no obvious parameter according to which this can be done. This paper looks at a new, efficient approach to determine the number of clusters.