We present an advanced method for feature extraction and cluster analysis of surface electromyograms (EMG) and acceleration signals in Parkinson's disease (PD). In the method, 12 different EMG and acceleration signal features are extracted and used to form high-dimensional feature vectors. The dimensionality of these vectors is then reduced by using the principal component approach. Finally, the cluster analysis of feature vectors is performed in a low-dimensional eigenspace. The method was tested with EMG and acceleration data of 42 patients with PD and 59 healthy controls. The obtained discrimination between patients and controls was promising. According to clustering results, one cluster contained 90% of the healthy controls and two other clusters 76% of the patients. Seven patients with severe motor dysfunctions were distinguished in one of the patient clusters. In the future, the clinical value of the method should be further evaluated.