Artificial Neural Networks (ANNs) have now been applied to a wide variety of real world problems in many fields of application. The attractive and flexible characteristics of ANNs make them well suited to the analysis of biological and medical signals especially in the brain signal analysis. The present study uses wavelet transform, with Haar as a basis function up to two level, as a data compression with ANN specifically on neurocomputing method called Independent Component Analysis (ICA) for analyzing the Magnetoencephalography (MEG) signals. The properties of wavelet transform such as locality, multiresolution and compression have led to a powerful approach to statistical signal processing. These signals have multichannel recordings of the magnetic fields, emerging from neural currents in the brain, and generate large amounts of data. Analysis of MEG recordings includes extracting the essential features of the neuromagnetic signals in the presence of artifacts.This study focuses on a signal processing technique, Independent Component Analysis (ICA), which is a simple and powerful method for analyzing multivariant data; it is a technique for the extraction of statistically independent components from a set of measured signals. The ICA algorithm is based on the assumption that the measured signals from a system are generated by linear, noiseless mixing of a set of underlying statistically independent sources. The algorithm performs a blind separation of the sources, commonly using techniques involving higher order statistics. In most cases, neurobiological data contain a lot of sensory noise, and the number of independent components is unknown. In this study, the ICA is applied to separate independent components data without knowing its number. The ICA is used on cases selected from continuous 122 – channel human agnetoencephalographic data. Both a nonlinear principal component analysis subspace (NPCA) learning rule and Bigradient algorithm are used to separate the MEG sources from biomedical sensor signals. The present model uses a feed forward neural network with three hidden layers: whitening, separating and estimating to the measured MEG signals. The neural network is trained with ICA algorithms (NPCA and Bigradient). The separated signals are then examined in terms of their order by using Euclidean technique, in terms of their independence by using RANK function, and in terms of nongaussinity by using kurtosis method. The results indicate clearly that this technique can be successfully applied to MEG data.