The aim of this study was to compare the clinical effectiveness of robot-assisted therapy with that of conventional occupational therapy according to the onset and severity of stroke. In this multicenter randomized controlled trial, stroke patients were randomized (1:1) to receive robot-assisted therapy or conventional occupational therapy. The robot-assisted training group received 30 min of robot-assisted therapy twice and 30 min of conventional occupational therapy daily, while the conventional therapy group received 90 min of occupational therapy. Therapy was conducted 5 days/week for 4 weeks. The primary outcome was the Wolf Motor Function Test (WMFT) score after 4 and 8 weeks of therapy. Overall, 113 and 115 patients received robot-assisted and conventional therapy, respectively. The WMFT score after robot-assisted therapy was not significantly better than that after conventional therapy, but there were significant improvements in the Motricity Index (trunk) and the Fugl-Meyer Assessment. After robot-assisted therapy, wrist strength significantly improved in the subacute or moderate-severity group of stroke patients. Robot-assisted therapy improved the upper-limb functions and activities of daily living (ADL) performance as much as conventional occupational therapy. In particular, it showed signs of more therapeutic effectiveness in the subacute stage or moderate-severity group.