To reduce boiler fuel cost, a fuel cost model is developed and applied to economic load dispatch (ELD). The purpose of ELD is to determine the power demands of individual generators so that total fuel cost of those generators may be minimized. In evaluating the fuel cost of each generator, an approximate model, which describes the fuel cost as a function of the generation, is used. Therefore, the accuracy of the model has a great influence on the preciseness of the minimization. So far, one of the most prevailing models has been a quadratic model, which describes the fuel cost as a quadratic function of the generation. However, this model often deviates from the actual data, especially during the time while the generation is dynamically varying. This fact implies that, if a model, which is accurate even when the generation dynamically varies, is developed and the model is applied to ELD, the total fuel cost can be reduced. First, taking into account that the deviation might be due to the dynamics of the automatic boiler controller (ABC), we employed as a supplementary model the autoregressive moving average (ARMA) model, which is capable of describing dynamics and is suitable for an on-line ELD calculation. Second, the ELD problem based on the ARMA-model-supplemented quadratic model is formulated as a quadratic programming (QP) problem, which can be solved by the standard QP method. Finally, the cost reduction effect is evaluated by computer simulations on Kansai Electric Power System.