Rehabilitation interventions that employ brain-computer interfaces (BCIs) require numerous sessions spanning over weeks or even months. Motor imagery (MI) based BCI models need to be calibrated at the start of each session with up to 20–30 min of data to achieve acceptable performance. Minimizing calibration times is important to increase the effective therapy time. Transfer learning could be used to address this issue, however, its performance with multi-session data in neurorehabilitation settings remains relatively underexplored. Here, transfer learning was applied to MI data from 1 naïve individual with SCI, which was combined with previous sessions data to reduce the calibration time needed at the beginning of each session. The use of transfer learning resulted in classification accuracy similar to standard calibration, but significantly reduced the amount of data required to be collected, from 18.06 ± 0.6 min to 3.84 ± 0.01 min (p < 0.01). Future research further improve transfer learning approaches to increase BCI classification accuracy.