The discibility matrix is one of the most important approaches to computing positive region, reduct, core and value reduct in rough sets. The subject of this paper is to develop a parallel approach of it, called tree expression. Its computational complexity for positive region and reduct is O(m2×n) instead of O(m×n2) in discibility-matrix-based approach, and is not over O(n2) for other concepts in rough sets, where m and n are the numbers of attributes and objects respectively in a given dataset (also called an information system in rough sets). This approach suits information systems with n >m and containing over one million objects.