Despite being a promising tool for establishing direct communication and control from the brain over external effectors, non-invasive Brain-computer interface (BCI) systems yield suboptimal performance in ~30% of users. Measuring the dynamical features that are relevant to the execution of a task to improve BCI performance remains an open challenge. Functional imaging is dominated by aperiodic and scale-free perturbations called “Neuronal Avalanches”, spreading across the whole brain. The sequence of regions recruited by avalanches could convey the processes underpinning BCI performance. To test our hypothesis, we used source-reconstructed magnetoencephalography signals in a BCI framework, where 20 healthy subjects were compared in resting-state and while performing a motor imagery (MI) task, in order to track the dynamical features related to motor imagery. Each signal was z-scored over time. For each condition, we estimated the avalanche transition matrix (ATM), containing the probability that regions j would be active at time t+1, given region i was active at time t. We computed the difference between the ATMs obtained from the two conditions in each subject, and validated them via permutation analysis. Then, we correlated the difference of the probabilities to BCI performance. All the significantly different edges cluster upon the premotor areas, involved during the MI task, and the cunei, involved during visual processing. The differences in the probabilities associated with edges incident upon areas such as the right paracentral lobule and the caudal middle frontal bilaterally directly correlate with the BCI scores. Our results suggest that avalanches capture functionally-relevant processes which are of interest for alternative BCI designing. Future work will consist of investigating the use of avalanche transition matrices as potential alternative features for classification in BCI.