The safety condition of a rock-fill dam has been evaluated by comparing the data of the measurement apparatuses with results getting from numerical analyses. Being surveyed the latest trends on the predicting method, the method (mainly numerical analysis) has been highly developed. However, its method has not always reflected accurately the real movement. The object of this report is to develop a new method combining the numerical analysis, the genetic algorithm and the artificial neural network. This new method could be successfully applied to the evaluation and prediction of pore pressure change in real two rock-fill dams.