Abstract: The variable selection is foundation in modeling of neural network.There are many influencing factors in thermal power generating units,which affect the coal consumption rate.If all sorts of influencing factors are included in input variable,related input variables will become overabundance,increase training burden of neural network,enhance possibility to get in local minimum point and reduce prediction precision of neural network.The paper firstly analyzed the factors that affected the standard coal consumption rate of thermal power generating units and proposed a method of variable selection based on sensitivity analysis,then it used the method to calculate contributing rate of each factor for output,and according to each contributing rate,the method selected six factors which had the most contributing rate from many influencing factors as input variable of neural network model.The simulation result showed that the method of variable selection could simplify structure of neural network,reduce training time of neural network and improve prediction precision.