Large databases are now available and many works offer to mine for the information within. Nevertheless, extracting interesting sites from biological sequential databases remains a challenging task, as they are known to occur with errors. Sequential patterns, by their flexible structure, enable to overcome this problem. However, those patterns are usually numerous while many of them are not relevant. Therefore, in this paper, we propose a new approach to mine for statistically significant sequential patterns. It extracts fewer patterns than more traditional approaches. Experiments show that such sequential patterns are pertinent for biological databases.