Objective: Independent of conventional neurofeedback training, in this study, we propose a tactile sensation assisted motor imagery training (SA-MI Training) approach to improve the performance of MI-based BCI. Methods: Twenty-six subjects were recruited and randomly divided into a Training-Group and a Control-Group. All subjects were required to perform three blocks of MI tasks. In the Training-Group, during the second block (SA-MI Training block), tactile stimulation was applied to the left or right wrist while the subjects were performing the left or right-hand MI task, while during the first block (Pre-Training block) and the third block (Post-Training block), subjects performed pure MI tasks without the tactile sensation assistance. In contrast, in the Control-Group, subjects performed the left and right-hand MI tasks in all three blocks. Results: The performance of the Post-Training block (83.2 $\pm$ 11.4%) was significantly (p = 0.0014) higher than that of the Pre-Training block (73.2 $\pm$ 16.3%). By contrast, in the Control-Group, no significant difference was found among the three blocks. Moreover, after the SA-MI Training, the motor-related cortex activation (i.e., ERD/ERS) and the R$^{2}$ coefficient in the alpha-beta band were enhanced, while no training effect was found in the Control-Group. Conclusion: The proposed SA-MI Training approach can significantly improve the performance of MI, which provides a novel training framework for MI-based BCI. Significance: It may be especially beneficial to those who are with difficulty in convention neurofeedback training or performing pure MI mental tasks to gain BCI control.