We present a filter bank (FB) approach to perform independent component analysis for adaptive noise cancelling. This approach is based on FBs, and its decimation provides much less computational complexity and faster convergence speed than the time-domain approach. In addition, the approach does not have a performance limitation unlike the frequency-domain approach. One can select the number of filters in the FB regardless of reverberation and implement the method to fit for parallel processing. We verify the effectiveness of the FB approach through simulations on adaptive noise cancelling. (C) 2003 Elsevier B.V. All rights reserved.