Use of published organizational data for a variety of purposes has the chance of violation of leakage of individual secret information. Preliminary efforts in this direction are susceptible to leakage of valuable information through quasi identifiers. Over the past few years, several algorithms based upon the concept of k-anonymity [1-2] have been developed, to handle such problems. A better privacy model, called l -diversity [3] was proposed to handle some of the problems in k-anonymity. Our main contribution in this paper is to improve the clustering phase of the OKA algorithm [4] so that it takes care of k-anonymity and l -diversity to a considerable extent and in combination with the improved second and third phases of the algorithm in [5] leads to an efficient l -diversity algorithm. We also show that all the three stages of the algorithm are necessary in order to cover different situations. © Springer-Verlag 2011.