This work proposes a system to predict the outcomes of the defibrillation during the period of ventricular fibrillation. Accurate outcomes can avoid inefficient defibrillation that causes severe myocardial injury. In this system, we apply a neural network model and use the frequency components of ECG signals as training data to determine the neuron coefficients. The trained system is then validated (tested) using different set of data to justify the performance. Experimental results are provided to show superior performance of the proposed system.