Subband adaptive filters have been proposed to avoid the drawbacks of slow convergence and high computational complexity associated with time domain adaptive filters. While the computational complexity is reduced, other undesired properties, such as signal delays and signal aliasing, are introduced. Aliasing effects may result in loss of perception in speech applications. A method for the design of oversampled filter banks is proposed to reduce these effects. The design method aims at reducing the inband aliasing as well as the reconstruction aliasing for the reason of achieving robustness when weighting in the subbands alters the subband signal phase and magnitude.