Injection attacks (e.g. XSS or SQL) are ranked at the first place in world-wide lists (e.g. MITRE and OWASP). These types of attacks can be easily obfuscated. Therefore it is difficult or even impossible to provide a reliable signature for firewalls that will detect such attacks. In this paper, we have proposed an innovative method for modelling the normal behaviour of web applications. The model is based on information obtained from HTTP requests generated by a client to a web server. We have evaluated our method on CSIC 2010 HTTP Dataset achieving satisfactory results.