Abstract OBJECTIVE. A subgroup of patients benefiting most from robot-assisted therapy (RT) has not yet been described. We examined the predictors of improved outcomes after RT. METHOD. Sixty-six patients with stroke receiving RT were analyzed. The outcome measures were the Fugl-Meyer Assessment (FMA), Wolf Motor Function Test (WMFT), Motor Activity Log (MAL), and Stroke Impact Scale (SIS). The potential predictors were age, side of lesion, time since onset, Modified Ashworth Scale (MAS) scores, accelerometer data, Box and Block Test (BBT) scores, and kinematic parameters. RESULTS. BBT scores were predictive of FMA (29%) and MAL (9%−15%) improvements. Reduced shoulder flexion synergy, as measured by less shoulder abduction during forward reach, and MAS–distal were predictive of WMFT–function improvements. MAS–distal was predictive of SIS–physical improvements. Demographic variables did not predict outcomes. CONCLUSION. Manual dexterity was a valuable predictor of motor impairment and daily function after RT. Outcomes at different levels may have different predictors.