When adjusting the borler combustion, the borler efficiency need to be constantly monitored.The traditional method of calculating boiler efficiency is complex.Based on the heat balance method,the main factors of influencing boiler efficiency was analysed deeply and the artificial neural network on-line monitoring model of boiler efficiency was established to predict boiler efficiency accurately and constantly in this paper. After precise analysis and tracking, the input variable for the artificial neural network on-line monitoring model of boiler efficiency was selected, so as to avoid larger error caused by the rough selection of input variable in the previous artificial neural network. At last,based on a 600MW boiler,the borler efficiency was predicted in this paper.we can easily know from the prediction result that the artificial neural network on-line monitoring model of boiler efficiency can predict the boiler efficiency accurately and constantly at a wide range condition.