Studying the behavior and location of neural activity using noninvasive method is an important approach for understanding the high order functions of human brain.MEG (magneto encephalography) can detect the magnetic fields which is originally generated by the intercellular or extracellular currents of neurons.Since the magnetic fields doesnt distort when it pass through the tissue and skull,hence MEG also has a higher spatial resolution (e.g.in source localization).The main difficulty comes from MEG is that the effect of noises since the magnetic fields of brain signal is relatively weak (50 to 1000 femto Tesla),especially for unaveraged single-trial data.To remove the additive noises and other interferences,or to find the behavior and location of source signals,the most popular and reliable method is to take an average over a large number of stimulus trials.Treating MEG data by taking the average,the signal-noise ratio (SNR) is increased.However,some important information in MEG data such as the strength of an evoked response and its dynamics will be lost.To improve effectiveness of MEG data analysis,this paper presents a novel method base on ICA (Independent Component Analysis) for the single-trial phantom data analysis[1].