Objective. To evaluate the clinical efficacy of BCI-supported mental practice and to reveal specific cognitive impairment which determine mental practice ineffectiveness and inability to perform MI. Material and methods. Fifty-five hemiplegic patients after first-time stroke (median age 54. 0 [44.0; 61.0], time from onset 6.0 [3.0; 13.0] month) were randomized into two groups — BCI and sham-controlled. Severity of arm paresis was measured by Fugl-Meyer Assessment of Motor Recovery after Stroke (FMA) and Action Research Arm Test (ARAT). Twelve patients from the BCI group were examined using neuropsychological testing. After assessment, patients were trained to imagine kinesthetically a movement under control of BCI with the feedback presented via an exoskeleton. Patients underwent 12 training sessions lasting up to 30 min. In the end of the study, the scores on movement scales, electroencephalographic results obtained during training sessions were analyzed and compared to the results of neuropsychological testing. Results. Evaluation of the UL clinical assessments indicated that both groups improved on ARAT and FMA (sections A—D, H, I) but only the BCI group showed an improvement in the ARAT’s grasp score (p=0.012), pinch score (p=0.012), gross movement score (p=0,002). The significant correlation was revealed between particular neuropsychological tests (Taylor Figure test, choice reaction test, Head test) and online accuracy rate. Conclusion. These results suggest that adding BCI control to exoskeleton-assisted physical therapy can improve post-stroke rehabilitation outcomes. Neuropsychological testing can be used for screening before mental practice admission and promote personalized rehabilitation.