To characterize the time-course of change in motor impairment, we examined voluntary elbow movement in stroke survivors over a period of one year post-stroke. We quantified several kinematic features of voluntary rapid elbow extension, by measuring the movement trajectory and its derivatives. The subjects were examined five times, at 1-, 2-, 3-, 6- and 12-months post-stroke. The data analyses had two steps. First we used the "growth mixture" model to characterize the recovery patterns of these kinematic parameters. Based on the observed measurements over 1 year, we found two classes of recovery patterns for each kinematic parameter. Subjects in class 1 started with a low value for each parameter and these values increased over time, while subjects in class 2 tended to start with higher value and showed widely divergent recovery patterns. Second, we used the logistic regression analysis to predict these recovery patterns based on Fugl Mayer Scale (FMS) of upper extremity measured on the first visit (i.e. 1 month after stroke). Based on the clinical evaluation of motor function (i.e. FMS) within the first month after stroke, these findings enable us to predict the recovery of arm impaired voluntary movement in hemiparetic stroke subjects at different times during the first year, and potentially beyond.