Original three-component polarimetric decomposition always suffers from some deficiencies such as negative power and scattering mechanism ambiguity. In fact, it is probably because of the inconsistency between the assumed models and PolSAR data. In this paper, we intend to solve the scattering mechanism ambiguity in original three-component decomposition, and later we will try to find out some solutions aiming at alleviating the negative power prob- lem. Considering that the nonnegative eigenvalue decomposition (NNED) is only applicative for the pixels that volume scattering dominates (the details will be discussed in Section 1), an improved model-based scheme is proposed, in which the decomposition will be hierarchical. If volume scattering is dominant, a revised NNED is adopted, for which the volume scattering employs the model proposed by Yamaguchi, and if volume scattering is not dominant, a revised four-component decomposition is adopted, for which the volume scattering employs the model proposed by Freeman. The entropy and alpha parameter are employed to identify the dominant scattering mechanism. Through this revision, the decomposition will be more accurate and the dominant scattering mechanism in each pixel will be preserved better. To verify and demonstrate the performance and advantages of the proposed decomposition, the L-band airborne AIRSAR San Francisco data and E-SAR Oberpfaffenhofen polarimetric data are used. Comparison studies are also carried out to show the improvements by the proposed method especially for the oriented urban areas.