Users identification by means of their keystroke pattern is an old and known technique. Several works had analysed and solved some of the most important issues in this area, but with the advances of technology, previous techniques have quickly become obsolete. Users as well as attackers spend a lot of time typing on their computers, locally and remotely. Their keystrokes leave a trace of patterns whose dynamism can be analysed and used to verify their identity. We propose to use unsupervised clustering algorithms to group user sessions together in order to correctly identify them. Furthermore, the verification of keystroke dynamics techniques has always been difficult because of the lack of a labeled free text dataset. To overcome this issue, we capture a large dataset of labeled... keystrokes of more than two and a half million digraphs. Results show that users can be accurately grouped.