This paper presents a video based method to quantify drug induced dyskinesias in Parkinson's disease (PD) patients. Dyskinetic movement in standard clinical videos of patients is analyzed by tracking landmark points on the video frames using non-rigid image registration. The novel application of Point Distribution Models (PDM) allows geometric variations and covariations of the landmark points to be captured from each video sequence. The PDM parameters represent quantifiable information that can be used to rate dyskinesia effectively, analogously to a neurologist's strategy of assessing the movement of multiple body parts simultaneously to effectively rate dyskinesia. A heuristic decision function is then developed using the PDM parameters to quantify the severity of the dyskinesia. The severity score using our decision function showed a high correlation to the dyskinesia rating of a neurologist on the corresponding patient videos.