Given that the manual feature selection process is cumbersome and not sufficiently accurate, a classification method of Power Quality Disturbance (PQD) based on a Gramian Angular Field (GAF) and a Convolutional Neural Network (CNN) is proposed when designing the power quality disturbance classifier. First, one-dimensional power quality disturbance signals are mapped to two-dimensional images. Then a network framework suitable for power quality disturbance classification is constructed based on the existing neural network. Finally, two-dimensional images are taken as input, and the CNN will automatically extract features from the massive disturbance samples and classify them. Simulation results show that this method has good classification performance in noisy data, and it is an effective power quality disturbance classification method. This work is supported by the National Natural Science Foundation of China (No. 61871133).