In this working paper, we study user identification via mouse movement. Instead of treating the problem as a multi-class classification task, we cast user identification as a one-class problem and propose to learn an individual model for every user. Preliminary empirical results show that our approach works for some but not all users. We report on lessons learned.