This paper proposes new transform-domain (TD) adaptive algorithms for long order adaptive filters. The scheme is based on decomposing the long adaptive filter into smaller subfilters. Each subfilter uses a different error signal in its update equation thus providing reduced coupling between subfilters and consequent improvement in convergence speed. We consider in this paper least mean-square adaptation of each subfilter along with power normalization, and derive an analytical model for the mean-square performance of the algorithm. Analysis assumes correlated Gaussian input data and will show that gains in convergence speed come at the expense of high misadjustment. A hybrid of the proposed algorithm and the TDLMS is described that offers significant performance improvements since it can be tuned to achieve very fast convergence while allowing low misadjustment. Performance advantages of the hybrid algorithm and theoretical results are verified by simulation examples.