The power of β oscillations is an essential pathological biomarker for movement disorders, parkinsonism in particular. Motor imagery training was reported to support self-regulate such β oscillations. Past studies had focused on the modulation of β oscillatory power per se, ignoring the intrinsic oscillatory characteristics-the nonlinearity of the waveform. This work applied ensemble empirical mode decomposition to decompose neural activities in multiple frequency bands without destroying the temporal characteristics of the raw signal at all scales. We explored the dynamics of the degree of nonlinearity plus the averaged power across all periods and frequency bands of interest and tested how motor imagery may or may not induce nonlinearities under various frequency bands. With motor imagery, the degree of nonlinearity for the β activity is significantly suppressed referenced to that without, of note, and the average power fails to present significant differences between segments with and without motor imagery training. Our results indicate that the degree of nonlinearity is a complementary and vital biomarker as the average power for β oscillations, thereby providing theoretical support for the possible application in motor imagery therapy. Clinical Relevance- This suggests that motor imagery can suppress irregular patterns of β oscillations for healthy, and the degree of nonlinearity is an effective feature in improving classification in training states for the MI-neurofeedback paradigm compared to that of the averaged power.