A database construction method which is based on the integration of relational database and distributed file system is researched for a large sample data base for interpretation of remote sensing images. It analyzes the database's construction process, and discusses the key technical problems and solution method. Based on the sample database, it studies the application method of sample data, analyzes some kinds of using cases and explores the application mode of sample data. Individual region's massive sample data are selected for verifying the method and its efficiency. At the same time, it takes 10 first-level classes which are defined in the land cover classification system for example, to analyze the spatial distribution and density characteristics of all kinds of sample data. The results show that the method of database construction and management which is based on the integration of relational database and distributed file system is very effective and applicative for sample data's searching, analyzing and promoted application.