Abstract: Large coal-fired power plant boiler has high operating costs when using the flue gas denitration technology, and establishing an effective denitration cost model is the basis to implement economy optimization. On the basis of the operating data of a 660 MW coal-fired boiler, back propagation（BP） algorithm and least squares support vector machine（LSSVM） were combined to predict the denitration cost. BP was applied to choose the input variables to reduce the dimension and complexity of the model, and the variables screened out was used as the final input of LSSVM to establish the BP-LSSVM model of denitration cost. The model was compared with the simple LSSVM model, and the result revealed that the model complexity is decreased and the model generalization ability is enhanced through the BP variable selection.