Abstract:  As an important mathematic method to deal with knowledge fuzziness and knowledge uncertainty,rough
sets theory has been paid more and more attention and widely used in various fields such as data mining, attributes
reduction with decision tree one of the most essential application.Though decision tree algorithm of rough sets theo-
ry have obtained many essential achicvements,there arc still some important issues need to be further resolved，espe-
dally the problems reflected in the low-efficiency about knowledge. Nowadays, it becomes a common interest for us
to find how to enrich the knowledge about information system. However,Classical rough set theory assumed that the
information systems arc complete, but in real applications, many information systems arc incomplete or inconsistent
because of different reasons.Therefore, knowledge acquisition from inconsistent information systems is a hot re-
search.The more relaxed formulation of lower approximation in terms of controlled degree of overlap between sets
rather than the inclusion relation was introduced in the context of the variable precision rough sets model. So the
lower and upper approximations could be naturally interpreted in probabilistic terms, leading to generalized notions of
rough set approximations,which will expand the elements of the boundary regions to the positive regions of Pawlak
rough sets theory and improve the ability of obtaining the exception information, In this paper,the definitions about
consistent set and the revised definition arc discussed based on Pawlak rough sets thcory,and further redefined the
variable precision consistent set. An approach for constructing the decision tree based on consistent set is proposed.
The results show that the sequential branches along the decision tree will get reasonable rules,which is simpler than
the traditional decision tree.