Fetal ECG is an obligatory signal to recognize the physical conditions of the fetus but the fetal ECG extraction is a taxing task for the signal researchers for the reason that the intrusion present in the recorded signal is more. The signal is actually recorded from abdomen of the mother, hence the interference is more, and mainly the predominant source of hindrance is the maternal ECG present in abdomen signal. The maternal ECG present in the abdomen is identified by a mapping the maternal ECG to abdomen ECG. The feed forward back propagation network perform this mapping and have the potential of error minimization using Levenberg-Marquardt back propagation method which updates weight and bias values of the network training function. The radial basis function uses local receptive fields to perform mapping. This article evaluates the performances of back propagation network and radial basis function network on fetal ECG extraction. The result of the extraction is further improved by moving average filtering and wavelet denoising techniques. The extraction of both the methods is compared with variance parameter, where the radial basis function provide 14.06% less variance than the back propagation method and similarly the clarity of extraction is better than the back propagation network. © International Science Press.