In this study, we propose a gear fault identification method based on adaptive-noise complementary ensemble empirical mode decomposition to solve the problem associated with the identification of gear faults. Initially, we used a fiber Bragg grating to extract the gear vibration signals, and uniformized the spectrum of vibration signal by adaptively adding Gaussian white noise to eliminate the mode mixing caused by the empirical modal algorithm. Subsequently, we used the correlation coefficient and the kurtosis value to obtain comprehensive evaluation indexes for selecting the effective components and extracting the features of the effective components. Finally, we used a support vector machine to identify the gear faults. The experimental results denote that the proposed method can be used to effectively identify the states of gears, including normal, mild-wear, severe-wear, pitting, cracks, broken teeth. Furthermore, the gear state identification accuracy is more than 90%.