Aiming to the existing defects of traditional passive islanding detection methods, such as large non-detection zone, thresholds difficult to determine and the results easily affected by power quality disturbances, a novel passive islanding detecting method based on singular value decomposition (SVD) and neural network for distributed generation (DG) is proposed. Initially, the voltage at point of common coupling (PCC) and output current of inverter are processed through S-transform to derive harmonic amplitudes matrixes. Then, the feature vector is formed by applying SVD to the matrixes. Further, it is determined by BP neural network whether there is an islanding phenomenon. The simulation results show that the method is faster than the traditional passive methods in islanding detection and can still accurately detect islanding in the state of power equilibrium, and is not easily affected by power quality disturbances, owning high accuracy, reliability and practicability.