Protection techniques commonly used on information systems are regularly hacked into by malicious users. This problem is mainly due to the fact that the reliability of such systems is based on correct administration, and on system users' correct handling. A new protection policy consists in confirming authentication of a user by analyzing his behavior. Such protection has never been used so far, mainly because of its poor performances. We are proposing here a statistical and geometrical method in order to build an authenticity profile of the subjects to authenticate. It consists in discovering on monitored system, some measurable parameters whose values can help us to discriminate authenticate subjects. This method also provides the means to analyze and evaluate the efficiency of the authentication tool before developing it. Finally, we are presenting an experimental application of a user's authentication by means of typing speed recognition.