The theory of rough sets provides a valuable approach in artificial intelligence and data mining. Optimal scale selection and attribute reduction are meaningful problems in rough set theory. Many related studies have been done in recent years. However, optimal scale selection based on the fuzzy conditional entropy and attribute reduction in multi-scale fuzzy decision systems have not been discussed yet. Aiming at this problem, we develop a novel algorithm based on the fuzzy conditional entropy to determine the optimal scale in a multi-scale fuzzy decision system. An attribute reduction algorithm is also proposed to reduce the redundant attributes in a fuzzy decision system that contains the optimal scale. Experimental results demonstrate that the proposed algorithms achieve higher classification accuracy and efficiency.