Protein structure alignment provides an effective way to detect and analyze fold mechanism, evolutionary history and biological function of proteins. In this work, we introduced a novel method named SABIC for protein structure alignment based on the internal coordinates (i.e. bond lengths, bond angles and torsion angles) of structure representation. SABIC provides multi-alignments as the output, from which various aspects of structural similarities between proteins can be identified. The experimental results on benchmark datasets show that SABIC performs better than other popular algorithms, such as DALI, CE and SSM. Using a new defined mQ-score of alignment, SABIC performs consistently better in detecting structural classifications of proteins. In addition, we detected the extreme value distribution form statistics of mQ-score, and then the statistical significance P-value of alignment can be obtained simultaneously. The presented SABIC algorithm has been implemented in C(++) and the software is available ( http://www.aporc.org/doc/wiki/SABIC ).