Abstract Background While constraint-induced movement therapy (CIMT) is one of the most promising techniques for upper limb rehabilitation after stroke, it requires high residual function to start with. Robotic device, on the other hand, can provide intention-driven assistance and is proven capable to complement conventional therapy. However, with many robotic devices focus on more proximal joints like shoulder and elbow, recovery of hand and fingers functions have become a challenge. Here we propose the use of robotic device to assist hand and fingers functions training and we aim to evaluate the potential efficacy of intention-driven robot-assisted fingers training. Methods Participants (6 to 24 months post-stroke) were randomly assigned into two groups: robot-assisted (robot) and non-assisted (control) fingers training groups. Each participant underwent 20-session training. Action Research Arm Test (ARAT) was used as the primary outcome measure, while, Wolf Motor Function Test (WMFT) score, its functional tasks (WMFT-FT) sub-score, Fugl-Meyer Assessment (FMA), its shoulder and elbow (FMA-SE) sub-score, and finger individuation index (FII) served as secondary outcome measures. Results Nineteen patients completed the 20-session training (Trial Registration: HKClinicalTrials.com HKCTR-1554); eighteen of them came back for a 6-month follow-up. Significant improvements (p