A model with annual power generation as objective function was established to optimize cascaded reservoir operation by using adaptive genetic algorithm. In addition to conventional constraints of water level and discharges etc, the model took ecological flow demand (EFD) into consideration. The model was applied to the operation of two cascade reservoirs on Yalong River in Southwest China. The optimal operation scheme was obtained through genetic algorithm, where a maximum total power generation could be achieved and all the constraints including EFD were satisfied. However, this solution would bring 8% hydropower loss compared with the optimal operation without EFD. In view of this, the quantitative response relation between hydropower loss and ecological flow demand were studied. The trade-off was found at the point when 70% EFD was guaranteed.