Mental practice (MP) is usually provided in combination with other therapies, and new developments for neurofeedback to support MP have been made recently. The objectives of this study were to evaluate the effectiveness of MP and to investigate the intervention characteristics including neurofeedback that may affect treatment outcome. The Cochrane Central Register of Controlled Trials, PubMed, Embase, KoreaMed, Scopus, Web of Science, PEDro, and CIRRIE were searched from inception to March 2017 for randomized controlled trials to assess the effect of MP for upper limb rehabilitation after stroke. Fugl-Meyer Assessment (FMA) was used as the outcome measure for meta-analysis. Twenty-five trials met the inclusion criteria, and 15 trials were eligible for meta-analysis. Among the trials selected for meta-analysis, MP was added to conventional therapy in eight trials or to modified constraint-induced movement therapy in one trial. The other trials provided neurofeedback to support MP: MP-guided neuromuscular electrical stimulation (NMES) in four trials and MP-guided robot-assisted therapy (RAT) in two trials. MP added to conventional therapy resulted in significantly higher FMA gain than conventional therapy alone. MP-guided NMES showed superior result than conventional NMES as well. However, the FMA gain of MP-guided RAT was not significantly higher than RAT alone. We suggest that MP is an effective complementary therapy either given with neurofeedback or not. Neurofeedback applied to MP showed different results depending on the therapy provided. This study has limitations because of heterogeneity and inadequate quality of trials. Further research is requested.