Abstract Objectives To (1) determine whether acceleration metrics derived from monitoring outside of treatment are responsive to change in upper extremity (UE) function; and secondarily to (2) compare metric values during task-specific training and while in the free-living environment, and (3) establish metric associations with an in-clinic measure of movement capabilities. Design Before-after observational study. Setting Inpatient hospital (primary purpose); outpatient hospital (secondary purpose). Participants Individuals (n=8) with UE hemiparesis <30 days poststroke (primary purpose); individuals (n=27) with UE hemiparesis ≥6 months poststroke (secondary purpose). Intervention The inpatient sample was evaluated for UE movement capabilities and monitored with wrist-worn accelerometers for 22 hours outside of treatment before and after multiple sessions of task-specific training. The outpatient sample was evaluated for UE movement capabilities and monitored during a single session of task-specific training and the subsequent 22 hours outside clinical settings. Main Outcome Measures Action Research Arm Test (ARAT) and acceleration metrics quantified from accelerometer recordings. Results Five metrics improved in the inpatient sample, along with UE function as measured on the ARAT: use ratio, magnitude ratio, variation ratio, median paretic UE acceleration magnitude, and paretic UE acceleration variability. Metric values were greater during task-specific training than in the free-living environment, and each metric was strongly associated with ARAT score. Conclusions Multiple metrics that characterize different aspects of UE movement are responsive to change in function. Metric values are different during training than in the free-living environment, providing further evidence that what the paretic UE does in the clinic may not generalize to what it does in everyday life.