Mental-Imagery based Brain-Computer Interfaces (MI-BCIs) allow their users to send commands to a computer via their brain activity, measured while they are performing specific mental tasks. While very promising (e.g., assistive technologies for motor-disabled patients) MI-BCI remain barely used outside laboratories because of the difficulty encountered by users to control such systems. Indeed, although some users obtain very good control performance after training, a substantial proportion remains unable to reliably control an MI-BCI. This huge variability led the community to look for predictors of MI-BCI control ability. In this paper, we introduce two predictive models of MI-BCI performance, based on a dataset of 17 participants who had to learn to control an MI-BCI by performing 3 MI-tasks: mental rotation, left-hand motor imagery and mental subtraction, across 6 sessions. These models include aspects of participants' personality and cognitive profiles, assessed by questionnaires. Both models, which explain more than 96% and 80% of MI-BCI performance variance, allowed us to define user profiles that could be associated with good BCI performances.