Adaptive filtering in subbands is a new technique for the real-time identification of large impulse responses like the ones encountered in acoustic echo cancellation. This technique generally allows computational savings as well as better convergence behavior. We give first an exact analysis of the critically subsampled two band modelization scheme. We demonstrate that adaptive cross-filters between the subbands are necessary for modelization with small output errors; moreover, we show that perfect reconstruction filter banks can yield exact modelization. We extend those results to the critically subsampled multiband schemes, and we show that important computational savings can be achieved by using good quality filter banks. Then we consider the problem of adaptive identification in critically subsampled subbands, and we derive an appropriate adaptation algorithm. We give a detailed analysis of the computational complexity of all the discussed schemes, and we verify experimentally the theoretical results that we have obtained. Finally, we discuss the adaptive behavior of the subband schemes that we have tested. We generally observe some degradation of the convergence performance in comparison with conventional schemes; however, the overall performance could be acceptable in practical use.