We consider the clustering problem with arbitrary shapes and different densities both within and between the clusters, where the number of clusters is unkno wn. We propose a new density-based approach in the graph theory context. The proposed algorithm has three phases. The first phase makes use of graph-based and density-based clustering approaches in order to identify the neighborhood structure of data points . The second phase detects outliers using the local o utlier concept. In the third phase, a hiearchical agglomeration is performed to form the final cluste rs. The algorithm is tested on a number data sets a nd compared with the well-known clustering algorithms in the literature. Its strengths and limitations ar e explored in detail.