As for physical data organization in data warehouse, there are mainly two kinds of methods, relational and multi-dimensional. These two methods have their own advantages and disadvantages, but as to improve the performance of OLAP (online analytical processing) query processing, the method of multi-dimensional array is superior. To solve the current problems in the method of multi-dimensional array, an improved multi-dimensional storage structure for data warehouse is proposed, and the concepts of logical storage and phtsical storage for data cube are given.According to this proposal,the original multi-dimensional data space is divided into many logical blocks,and a logical block is divided into many physical blocks.This multi-dimensional storage structure takes the characteristics of the large amount and highly sparse multi-dimensional array into consideration fully,and a new distributing and compressing method for the multi-dimensional array is adopted.These availably solve efficiency problems of the aggregation query along with the inner level of the dimension and query,and dramatically improve the response time of the aggregation query.In particular,these methods also bring additional b9enefit for incremental maintenance of the multi-dimensional array.