In view of the shortcomings of existing formal concept attribute reduction algorithms, such as the problems that the time complexity of attribute reduction is too high, redundancy calculation exists in attribute and attribute value comparison, and storage overhead is great and so on, by combining the idea of granular computing this paper proposes a conceptual attribute reduction model based on attribute classification. Firstly, two operators are defined to classify the relations among attributes. Then, a reduction rule is established by attribute classification relation. On this basis, this paper proposes a formal concept reduction algorithm based on attribute classification, which reduces the redundancy calculation and storage overhead while keeping the current minimum time complexity constant, which improves the computational efficiency of attribute reduction. Finally, the validity of the formal attribute reduction algorithm based on classification is verified by examples and simulation experiments.