An attribute reduction algorithm of rough set is put forward on the ba sis of the analysis of bearing vibration signals. The worst Fisher criterion was adopted as heuristic information to improve the searching efficiency and the maxmin ant system was selected for the condition attribute reduction in the comp o und fault of a bearing in a gearbox. Considering the classification accuracy and the number of the condition attribute, the fault diagnosis decision table was s implified by using the rough sets theory. The results show that the method can m arkedly raise the efficiency of bearing fault diagnosis.