A novel multi-sensor-based method to detect and classify short circuit faults in radial power distribution systems, including the effects of regulators and distribution transformers, is proposed in this work. This new scheme first calculates the correlation among the data of different sensors and uses principal component analysis (PCA) to reduce the data dimensions. Then, kernel support vector machine (SVM) classifiers are applied to detect faults and identify faulty phases. The proposed method is simulated and tested for normal operations like load switching and different types of faults under different signal-to-noise ratio (SNR) scenarios with various fault durations and fault impedances. The Gaussian mixture noise model is used in the simulations to test the robustness of the algorithm. Two distribution system models (an unbalanced feeder and a balanced feeder) are used in this work to determine the impacts of system configuration on the proposed method. Finally, relationships among the number of sensors, sampling rate of the sensors, and detection performance are discussed.