To detect the imagined limb movement from EEG for the use in BCI, the increase (ERS) and decrease (ERD) of the band power of the EEG originated from the sensorimotor cortex are commonly used. A spatial filter using neighboring channels is generally applied to the measured EEG for detecting such brain activity related to the motor imagery. However, the configuration and location of the spatial filter have been selected by the empirical method on trial-and-error basis. In this study, we recorded the EEG during motor imagery of left hand, right hand and feet from five subjects, and the ICA (independent component analysis) was applied to discover the spatial filters for extracting event-related EEG components of the motor imagery. It was suggested that the application of ICA might offer the experimenters appropriate local spatial filters, or at least, the initial guess for designing or selecting custom local spatial filters.