In traditional cost-sensitive attribute reduction, the variation of decision cost is referred to as a global difference of costs because the considered decision cost is the variation of sum of decision costs over all objects. However, such reduction does not take the variation of decision costs of each object into account. To solve this problem, a local view based cost-sensitive attribute reduction is introduced. Firstly, through considering the variation of decision costs of single object if the used attributes change, a local difference of costs is presented. Secondly, on the basis of the fuzzy decision-theoretic rough set model, a new significance function is given to measure the importance of attribute. Finally, the experimental results illustrate that by comparing the traditional reduction, the proposed local view can decreases both global and local differences of costs effectively on several UCI data sets,