To examine the responsiveness and validity of the Action Research Arm Test (ARAT) in a population of subjects with mild-to-moderate hemiparesis within the first few months after stroke. Data were collected as part of the Very Early Constraint-Induced Therapy for Recovery from Stroke trial, an acute, single-blind randomized controlled trial of constraint-induced movement therapy. Subjects were studied at baseline (day 0), after treatment (day 14), and after 90 days (day 90) poststroke. Inpatient rehabilitation hospital; follow-up 3 months poststroke. Fifty hemiparetic subjects. Not applicable. At each time point, subjects were tested on: (1) the ARAT, (2) clinical measures of sensorimotor impairments, (3) in the kinematics laboratory where they performed reach and grasp movements, and (4) clinical measures of disability. Blinded raters performed all evaluations. Analyses at each time point included calculating effect size as indicators of responsiveness, and correlation and regression analyses to examine relationships between ARAT scores and other measures. The ARAT is responsive to change, with effect sizes greater than 1.0 and responsiveness ratios of 7.0 at 3 months poststroke. ARAT scores were related to sensorimotor impairment measures, 3-dimensional kinematic measures of movement performance, and disability measures at all 3 time points. The ARAT is a responsive and valid measure of upper-extremity functional limitation and therefore may be an appropriate measure for use in acute upper-extremity rehabilitation trials.