As a generalization of rough set, fuzzy rough set plays a significant role in attribute reduction. Reduct size and dependency degree are two important evaluation criteria to measure reduction performance. Traditional reduction algorithms are mainly designed with the direction of maximizing dependency degree and reduct size is not taken into consideration. This paper emphasizes the significance of reduct size and proposes an multiobjective difference algorithm based on neighborhood mutation information, which makes reduct size a separate objective in attribute reduction. Hence the reduction problem is turned into multiobjective optimization problem, which considers reduction performance from aspects of both attribute number and dependency degree. By using goal-sequence domination scheme, the reduction performance can be stipulated from aspects of attribute number and dependency degree and satisfactory reduction results can be obtained. Several UCI data sets are applied to analyze the algorithm performance. The experimental results show that this method can obtain more comprehensive reduction results, which is feasible and effective.