Abstract: A method and system for optimizing a database query. A database table populated with data is received and scanned. Statistics and single column histograms associated with single columns of the table are determined. Cardinality based on the statistics and histograms is estimated. All possible correlations among multiple columns are determined by performing clustering data mining that partitions data in the table into clusters. Top ranked columns based on the correlations are determined. The difference between the estimated cardinality and a support count of a cluster is determined to exceed a threshold, and in response, multiple column histograms based on the top ranked columns are determined. An optimal query plan based on the multiple column histograms is generated.