Considering the poor accuracy of grey relational analysis method in fault diagnosis field, a rough set based weighted grey fault diagnosis algorithm is provided. The algorithm extracts the core characteristic parameters described the fault description and calculate their weight with rough sets and its attribute reduction method, the possible fault pattern of the unknown pattern is judged by the relation degree which calculated quantitatively with each typical fault pattern in history diagnosis record based on weighted grey relational analysis method. Experiment result of recognizing a diesel engine working state indicates that rough set based weighted grey fault diagnosis algorithm find effectively the optimization characteristics parameter for fault description, emphasize the importance of the different characteristics parameters, improve the accuracy of grey diagnose algorithm, and can play significant performance in actual application.