Background and Objective: This paper proposes a better and improved K-Means clustering algorithm which deals with datasets comprising of a small amount of labeled data objects and the rest containing unlabeled data. Materials and Methods: The labels provide a fair idea about the number of clusters that the data might amount to. Since majority of the dataset contains unlabeled data, the proposed algorithm also finds the initial cluster points dynamically, by finding data objects that are most dissimilar to the existing cluster points. Results and Conclusion: Thus this version of the K-Means strives to function using less number of iterations due to the availability of labeled information on one hand and also results in a fairly optimal number of clusters present in the given dataset. © 2016, International Journal of Pharmacy and Technology. All rights reserved.