Motor imagery (MI) training improves motor performance, but the inter-individual variability of this improvement remains still unexplored. In this study, we tested the influence of imagery ability on the performance improvement following MI training. Twenty participants were randomly distributed into the MI or control group. They actually performed, at pre- and post-test sessions, a revisited version of the Nine Hole Peg Test, a speed-accuracy trade-off task, commonly used in clinics. Between the tests, the MI group mentally trained on the task (5 blocks of 10 trials), while the control group watched a non-emotional documentary. Before and during MI training, we tested the imagery ability of the MI group, by the revised version of Movement Imagery Questionnaire and by the estimation of vividness for the movement task at each block (subjective evaluation-SE). In the post-test, the MI group significantly decreased the movement duration by -12.1±5.7% (P<0.001), whereas the control group did not (-2.68±5%, P=0.68). For the MI group, the percentage of improvement was correlated neither to the MIQ-R nor to the SE reported after block 1. However, we observed an evolution of the SE during training, with a positive correlation between performance improvement and SE at block 4 (R=0.61, P=0.03) and at block 5 (R=0.68, P=0.04). The current study shows that motor performance may be positively influenced, whilst not predicted, by the capacity to form vivid movement images throughout the mental training. These findings are of interest for clinical interventions using MI as a complementary rehabilitation tool.