Abstract: Sequence alignment was one of the most popular operations in bioinformatics. The key issue of alignment was how to improve matching speed in a large sequences database. In this study, a full-sensitivity algorithm was proposed to solve the problem of finding all local alignments over a given length w with an error rate at most e. The proposed algorithm was implemented on a q-gram index. First, a large part of irrelevant subsequences were eliminated quickly by effective filtrating with new diagonal features. These new diagonal features were extracted from match-regions by analyzing the edit matrix of query sequence and database. Second, the unfiltered regions were verified by smith-waterman algorithm to search the true matches. The experimental results demonstrate that the proposed algorithm improves the filtration efficiency in a short filtration time and the algorithm is always faster than the well-known SWIFT on condition of low max error rate. This result is of great practical to local alignment with low error rate and short window size.