Many people with stroke experience incomplete recoveries, leaving them with upper extremity (UE) deficits affecting their long-term independence. Interventions including virtual reality (VR) and robotics have been developed to foster neuroplasticity post stroke. Few of the many studies examining these interventions consider the impact of both timing and dosage. The primary aim of this randomized controlled trial was to investigate (1) dosage and (2) timing of UE VR/robotic training in the subacute period post stroke. 100 participants were consented 5-30 days after stroke. They were randomized to an Early (first month) or Delayed (second month) VR/robotic group (EVR/DVR), a dose matched usual care group (DMUC) or a usual care group (UC). Participants were evaluated using impairment, motor function, and quality of life measures immediately before, after, and 1 month after training, and 4 and 6 months post stroke. At 4 months post stroke the DVR group showed a higher rate of change from baseline on the Action Research Arm Test compared to the EVR group. This difference was not sustained; none of the training groups demonstrated significantly better scores on any outcome measure 6 months post stroke. Growth mixture modeling revealed three groups with patterns of recovery associated with early finger movement. At 6 months post stroke, the EuroQol was moderately correlated with impairment and activity.