This paper presents an approach to the multiple sequence alignment (MSA) problem by applying genetic algorithms with a reserve selection mechanism. MSA is one of the most fundamental operations in bioinformatics, which plays an important part in predicting the structure, function and evolution of biological sequences. In order to solve the MSA problem efficiently, genetic algorithms (GAs) were applied. As the number and length of sequences increase, however, GAs are usually suffered from premature convergence where they are easily trapped into local optima. In this paper, we employ the reserve selection that is a new selection scheme to avoid premature convergence in GAs. Empirical studies demonstrate that genetic algorithms with reserve selection (GARS) could bring about a rise in the quality of multiple sequence alignment when compared with standard GAs.