Khiops is an automatic supervised classification tool for mining large multi-tables databases. The predictive importance of input variables is evaluated by the mean of discretization models in the numerical case and of value grouping models in the categorical case. In the case of a multi-tables database, for exemple customers with their purchases, an analysis data table instances × variables is produced using automatic feature construction. The supervised classification model is a naive Bayes classifier, with variable selection and model averaging. The tool is designed for the analysis of large databases, with millions of instances, tens of thousands of variables and hundreds of millions of records in the secondary tables.