The aim of this study is to identify movement characteristics associated with motor fluctuations in patients with Parkinson?s disease by relying on wearable sensors. Improved methods of assessing longitudinal changes in Parkinson?s disease would enable optimization of treatment and maximization of patient function. We used eight accelerometers on the upper and lower limbs to monitor patients while they performed a set of standardized motor tasks. A video of the subjects was used by an expert to assign clinical scores. We focused on a motor complication referred to as dyskinesia, which is observed in association with medication intake. The sensor data were processed to extract a feature set responsive to the motor fluctuations. To assess the ability of accelerometers to capture the motor fluctuation patterns, the feature space was visualized using PCA and Sammon?s mapping. Clustering analysis revealed the existence of intermediate clusters that were observed when changes occurred in the severity of dyskinesia. We present quantitative evidence that these intermediate clusters are the result of the high sensitivity of the proposed technique to changes in the severity of dyskinesia observed during motor fluctuation cycles.