Rough set theory proposed by professor Pawlak is an important mean to solve the problem of uncertain boundary region. Pawlak constructed two crisp boundaries for the set with uncertainty boundary but did not give any exact or approximate methods of using the existing knowledge base to build an approximation set of a target concept .In order to solve this problem, in the previous researches a method for looking for this kind of approximation target concept (set) is proposed. However, that method does not give out a kind of optimal approximation set. In this paper, firstly, the concept of the similarity between the target set and its approximation set and the method for constructing approximation set of rough set are reviewed, and the operation properties are proposed and proved respectively. Secondly, an interval of λ is found, and in this interval R_λ(X) is more similar to the target concept X than the upper-approximation set R(X) or lower-approximation set R(X). Finally, the conditions of R_0.5(X) as an optimal approximation set of the target concept X are proposed.