Unplanned island detection will cause serious impact on the power grid and personal injury, so the implementation of the island detection must be accurate and rapid. Aiming at the current island detection technology based on signal processing, this paper proposes the novel island detection method based on wavelet packet energy entropy and BP neural network. Firstly, the common point voltage and the output current of an inverter are decomposed and reconstructed by wavelet packet transformation, and the reconstructed sequences are obtained. Then, entropy operation is conducted on the sequences, which can get a more stable and representative eigenvector. In this way, we can distinguish the energy distribution before and after island occurrence more effectively. Finally, the BP neural network is adopted to judge the island for realizing the island detection. In addition, the MATLAB/Simulink simulation shows the effectiveness of the proposed method, and the response speed is fast. And there is no need to introduce disturbances, which will not cause the power quality problems, and its stability is high, and the detection blind zone is small.