This paper presents a method for blind separation of auditory event-related brain responses into independent components using independent component analysis (ICA). The method decomposes averaged event-related potential (ERP) data recorded from the scalp into a sum of components with fixed scalp distributions and sparsely activated, maximally independent time courses. Unlike traditional methods such as dipole fitting or principal component analysis (PCA), ICA minimizes higher-order dependencies and does not model the locations of the generators in the head. The algorithm was applied to ERP data from an auditory vigilance experiment, where it successfully decomposed the major response peaks into one or more ICA components with relatively simple scalp distributions. Three components were active only when the subject detected the targets, three others only when the target went undetected, and one in both cases. Three additional components accounted for the steady-state brain response to a 39-Hz background click train. The decomposition proved robust across sessions and changes in sensor number and placement. The method allows for the comparison of responses from multiple stimuli, task conditions, and subject states. The ICA algorithm is based on an "infomax" neural network and uses stochastic gradient ascent to maximize the entropy of the data. It was implemented using a computationally efficient version of the algorithm to decompose brief evoked brain responses into temporally independent components. The results showed that ICA could accurately decompose components with skewed distributions even without specific nonlinearities. The method is computationally efficient and can be applied to data from a hundred or more channels. The ICA algorithm was used to decompose two 1-s ERPs into 14 ICA components, which were found to have distinct scalp distributions and time courses. The decomposition was stable across different sessions and electrode placements. The results demonstrated that ICA could effectively separate auditory event-related brain responses into independent components, providing a new and potentially useful approach for analyzing complex ERP data. The method may be particularly useful for comparing the activations of brain response components in different stimulus and cognitive task conditions. The ICA decomposition may also be useful for preprocessing data prior to applying source localization algorithms and for observing event-related changes in the spatial structure of correlated ongoing EEG activity in multiple brain areas. The method is also applicable to magnetoencephalographic (MEG) data and can be generalized to track changes in the spatial structure of EEG or MEG activity in different brain states.This paper presents a method for blind separation of auditory event-related brain responses into independent components using independent component analysis (ICA). The method decomposes averaged event-related potential (ERP) data recorded from the scalp into a sum of components with fixed scalp distributions and sparsely activated, maximally independent time courses. Unlike traditional methods such as dipole fitting or principal component analysis (PCA), ICA minimizes higher-order dependencies and does not model the locations of the generators in the head. The algorithm was applied to ERP data from an auditory vigilance experiment, where it successfully decomposed the major response peaks into one or more ICA components with relatively simple scalp distributions. Three components were active only when the subject detected the targets, three others only when the target went undetected, and one in both cases. Three additional components accounted for the steady-state brain response to a 39-Hz background click train. The decomposition proved robust across sessions and changes in sensor number and placement. The method allows for the comparison of responses from multiple stimuli, task conditions, and subject states. The ICA algorithm is based on an "infomax" neural network and uses stochastic gradient ascent to maximize the entropy of the data. It was implemented using a computationally efficient version of the algorithm to decompose brief evoked brain responses into temporally independent components. The results showed that ICA could accurately decompose components with skewed distributions even without specific nonlinearities. The method is computationally efficient and can be applied to data from a hundred or more channels. The ICA algorithm was used to decompose two 1-s ERPs into 14 ICA components, which were found to have distinct scalp distributions and time courses. The decomposition was stable across different sessions and electrode placements. The results demonstrated that ICA could effectively separate auditory event-related brain responses into independent components, providing a new and potentially useful approach for analyzing complex ERP data. The method may be particularly useful for comparing the activations of brain response components in different stimulus and cognitive task conditions. The ICA decomposition may also be useful for preprocessing data prior to applying source localization algorithms and for observing event-related changes in the spatial structure of correlated ongoing EEG activity in multiple brain areas. The method is also applicable to magnetoencephalographic (MEG) data and can be generalized to track changes in the spatial structure of EEG or MEG activity in different brain states.