While being very promising for a wide range of applications, Mental-Imagery based Brain-Computer Interfaces (MI-BCIs) remain barely used outside laboratories, notably due to the difficulties users encounter when attempting to control them. Indeed, 10 to 30% of users are unable to control MI-BCIs (so-called " BCI illiteracy ") while only a small proportion reach acceptable control abilities. This huge inter-user variability has led the community to investigate potential predictors of performance related to users' personality and cognitive profile. Based on a literature review, we propose a classification of these MI-BCI performance predictors into three categories representing high-level cognitive concepts: (1) users' relationship with the technology (including the notions of computer-anxiety and sense of agency), (2) attention and (3) spatial abilities. We detail these concepts and their neural correlates in order to better understand their relationship with MI-BCI user-training. Consequently, we propose, by way of future prospects, some guidelines to improve MI-BCI user-training.