Multiple sequence alignment is a broadly used methodology in biological applications. It is expected to locate consensus sequence stretches with evolutionary and functional conservation. However, when sequence similarity among the queries becomes low, it works poorly. The aim of this study is to incorporate important biological knowledge and assumption to improve the quality of a general alignment on low similarity sequences such as carbohydrate binding module (CBM) families. Since the recognition of characteristic patterns in CBMs does not apply to a general model, a more accurate scoring function employing secondary-structure-based and key-residue-weighted algorithms for alignment was designed to approach this goal. Our results indicated that the new method was practically applicable to identify the key residues in terms of three-dimensional structures, while conventional tools could fail. ; PRIB 2008 proceedings found at: http://dx.doi.org/10.1007/978-3-540-88436-1