Abstract Background and purpose There is little evidence available on the use of robot-assisted therapy in subacute stroke patients. A randomized controlled trial was carried out to evaluate the short-time efficacy of intensive robot-assisted therapy compared to usual physical therapy performed in the early phase after stroke onset. Methods Fifty-three subacute stroke patients at their first-ever stroke were enrolled 30 ± 7 days after the acute event and randomized into two groups, both exposed to standard therapy. Additional 30 sessions of robot-assisted therapy were provided to the Experimental Group. Additional 30 sessions of usual therapy were provided to the Control Group. The following impairment evaluations were performed at the beginning (T0), after 15 sessions (T1), and at the end of the treatment (T2): Fugl-Meyer Assessment Scale (FM), Modified Ashworth Scale-Shoulder (MAS-S), Modified Ashworth Scale-Elbow (MAS-E), Total Passive Range of Motion-Shoulder/Elbow (pROM), and Motricity Index (MI). Results Evidence of significant improvements in MAS-S (p = 0.004), MAS-E (p = 0.018) and pROM (p