In order to support fast response for ad-hoc and complex OLAP queries, a data cube approach is introduced. Among the various data cube methods proposed in the literature, quotient cube and QC-tree are two important ones, because they try to condense the size of a data cube, while keeping its semantics. However, the former does not store any semantics and the latter stores the semantics in an obscure and implicit manner. To follow this trend and solve the existing problem, drill-down cube is proposed in this paper. Drill-down cube considers the data cube store from the point of view of drill-down semantics, which stores the drill-down semantics between classes, not the content of classes. In a drill-down cube, each class is represented as a node and each direct drill-down relation is captured and represented as an edge between two nodes. The analysis and experiments show that drill-down cube not only reduces the storage size of a data cube dramatically, but also captures the drill-down semantics of the data cube naturally and clearly. The query answering against a drill-down cube, including both point queries and range queries, is also discussed. The key idea behind is to drill down to all target nodes from the root. The query answering of drill-down cube performs fairly well, especially for range queries, and this is verified in the empirical evaluation.