A novel method for discrimination of dynamic muscle contractions between patients with Parkinson's disease (PD) and healthy controls on the basis of surface electromyography (EMG) and acceleration measurements is presented. In this method, dynamic EMG and acceleration measurements are analyzed using nonlinear methods and wavelets. Ten parameters capturing Parkinson's disease (PD) characteristic features in the measured signals are extracted. Each parameter is computed as time-varying, and for elbow flexion and extension movements separately. For discrimination between subjects, the dimensionality of the feature vectors formed from these parameters is reduced using a principal component approach. The cluster analysis of the low-dimensional feature vectors is then performed for flexion and extension movements separately. The EMG and acceleration data measured from 49 patients with PD and 59 healthy controls are used for analysis. According to clustering results, the method could discriminate 80 % of patient extension movements from 87 % of control extension movements, and 73 % of patient flexion movements from 82 % of control flexion movements. The results show that dynamic EMG and acceleration measurements can be informative for assessing neuromuscular dysfunction in PD, and furthermore, they may help in the objective clinical assessment of the disease.