Traditional data warehouse mainly adopts methods of data organization of dual level of granularity, partitioning data and so on, which are department-oriented or OLAP and cannot satisfy the need of customized data analysis. The article proposes a method of data organization in data warehouse based on predicate classification through analyzing conditions (predicate) of former queries and statistics. The method, which bases on affinitive frequency structures an equivalence relation, generates relative materialized views through predicate classification and selection of core predicate in order to form customer-oriented method of data organization in data warehouse. In sum, this method improves the efficiency of data analysis for customers and provides a new quotable method for organizing data in data warehouse.