Abstract: For fault in control system being difficult to be detected and found in the early stage, signed directed graph (SDG) and qualitative trend analysis based fault diagnosis is proposed. The SDG model is built for fault diagnosis, and qualitative trend analysis is used to monitor the state of units in the control system, extract their trends. When fault occurs, inverse inference based on qualitative trends is used to find the real fault cause. This method is tested on the atmospheric tower unit. The results show that this method can detect and identify various faults inside control system (such as manual misoperation, controller failure and so on) and can be applied in the real industry.