Cluster analysis is an important technique for data mining and knowledge discovery in spatial databases. Its main advantage is the ability to find interesting structures or clusters directly form the spatial data without using any background knowledge. Some available algorithms are reviewed and a mathematical morphology based clustering algorithm (MMC) is presented in this paper. Clusters with arbitrary shape can be discovered by using MMC algorithm

and the optimal cluster number is automatically determined by a heuristic method. The algorithm can be implemented in vector databases as well as in raster databases. The experiments show that the new algorithm is feasible and effective for discovering clusters in spatial databases and is robust when clustering in databases with noise.