Abstract: As an effective technique of on-line measurement, soft sensor has been studied by many efforts. Being the premise of soft sensor, it is important to select secondary variables, which can directly influence the efficiency of soft sensor. Grey relation analysis gives a novel method to dig correlation between variables and is suitable to confirm the correlation degree between little-sample data in uncertain systems. Through a detailed analysis of secondary variables′ selection for soft sensor, a uniform incidence degree algorithm is presented based on grey relation theory to efficiently select secondary variables. And a practical method to eliminate redundant variables is also proposed. Meanwhile, a practical example is given to verify the feasibility and effectiveness of the method. Simulation results show that the method is efficient and helpful to enhance the accuracy of soft sensor.