Abstract: Clustering is a promising application technique for many fields including data mining, pattern recognition, image processing, compression and other business applications. DBSCAN is a density based clustering algorithm that can efficiently discover clusters of arbitrary shape and can effectively handle noise. However, it requires large volume of memory support and needs a lot of I/O costs when dealing with large scale databases because it operates directly on the entire databases. Furthermore, clustering quality will degrade when the cluster density and the distance between clusters are not even. In this paper, an improved DBSCAN algorithm is presented on the basis of data partitioning. Experimental results show that the new algorithm is superior to the original DBSCAN in efficiency.