Gait analysis is important in diagnosing and evaluating certain neurological diseases such as Parkinson's disease (PD). In this paper, we show the ability of our wireless inertial sensor system to characterize gait abnormalities in PD. We obtain physical features of pitch, roll, and yaw rotations of the foot during walking, use principal component analysis (PCA) to select features, and use the support vector machine (SVM) method to create a classification model. In the binary classification task of detecting the presence of PD by distinguishing between PD and control subjects, the model performs with over 93% sensitivity and specificity, and 97.7% precision. Using a cost-sensitive learner to reflect the different costs associated with misclassifying PD and control subjects, performance of 100% specificity and precision is achieved, while maintaining sensitivity of close to 89%. In the multi-class classification task of characterizing parkinsonian gait by distinguishing among PD with significant gait disturbance, PD with no significant gait disturbance, and control subjects, 91.7% class recall for control subjects is achieved and the model performs with 84.6% precision for PD subjects with significant gait disturbance. The features selected for this classification task indicate the features of gait that are principal in discriminating gait abnormalities due to PD compared to a normal gait. These results demonstrate the ability of our wireless inertial sensor system to successfully detect the presence of PD based on physical features of gait and to identify the specific features that characterize parkinsonian gait.