Hemiparesis, the weakness of one side of the body, affects the ability of stroke survivors to move and walk. It is generally diagnosed through motor tests performed as part of neurological examinations such as the NIH Stroke Scale (NIHSS), a subjective evaluation that requires the presence of an experienced neurologist. Here we report on an alternative way for computationally identifying hemiparesis that leverages body joint position data captured by the Microsoft Kinect. We employed support vector machines with 14 stroke subjects and 21 controls to characterize hemiparesis based on 4 core body angles recorded while the participants were simply sitting at rest, waiting for their neurologist. When comparing our results to neurologists' NIHSS scores, we were able to always identify right-side hemiparesis, left-side hemiparesis, or no hemiparesis using a leave-one-subject-out analysis. With additional data, our ultimate aim is to include the hemiparesis detection system presented here in a larger, multimodal tool that characterizes stroke based on several stroke-associated deficits. We envision deploying this tool in emergency settings for faster and more precise stroke severity assessments done in real-time.