Bradykinesia is one of the primary characteristic symptoms of Parkinson's disease (PD). Ten-second whole-hand-grasps action was chosen to assess bradykinesia severity in this study. A quantification assessment system based on a self-developed wearable device was proposed to assess the severity of the parkinsonian bradykinesia. The proposed assessment method used an attitude-estimation algorithm to extract the parkinsonian bradykinesia parameters. A regression model was adopted to fit the characteristic parameters with the clinical UPDRS ratings judged by neurologists. Clinical experiment with 15 PD patients and 5 age-matched healthy controls demonstrated that the predicted bradykinesia scores by proposed model correlated well with the judgments of neurologists (r2=0.99). The proposed quantification model demonstrated the greater goodness-of-fit compared with the related works.