An application of an artificial neural network (ANN) has been implemented in this article to model the nonlinear relationship of the harvested electrical power of a recently developed piezoelectric pendulum with respect to its resistive load R (L) and magnetic excitation frequency f. Prediction of harvested power for a wide range is a difficult task, because it increases dramatically when f gets closer to the natural frequency f (0) of the system. The neural model of the concerned system is designed upon the basis of a standard multi-layer network with a back propagation learning algorithm. Input data, termed input patterns, to present to the network and the respective output data, termed output patterns, describing desired network output that are carefully collected from the experiment under several conditions in order to train the developed network accurately. Results have indicated that the designed ANN is an effective means for predicting the harvested power of the piezoelectric harvester as functions of R (L) and f with a root mean square error of 6.65 x 10(-3) for training and 1.40 for different test conditions. Using the proposed approach, the harvested power can be estimated reasonably without tackling the difficulty of experimental studies and complexity of analytical formulas representing the concerned system.