With the widespread access of nonlinear loads, enterprises and consumers are faced with the problem of power quality disturbance (PQD) in the grid. The detection and classification of multiple PQDs are considered to be a challenging task. In this article, a novel hybrid method based on Stockwell transform (ST) and deep learning is proposed to detect and classify multiple PQDs (MPQDs). Compared with previous cases where only single or dual PQDs are considered, this article fully considers the existence of MPQDs compounds in real situations and designs a more general and automated approach based on deep learning for automated feature selection and classification. First, the S-transform is used to extract the features of the 1-D signal and take the modulus of the result to obtain the ST matrix of the signal. In order to improve the antinoise performance of the system, the parameters for generating the time–frequency-domain contour are optimized, so that contour images with less noise can be drawn. Second, the above images are fed into the designed convolutional neural network (CNN) for training. A total of 37 PQDs detection and classification tasks including single and multiple disturbances were completed. It is compared with other existing methods to demonstrate its robustness under noisy environments. Finally, an experimental platform for MPQDs was built to verify the proposed method, and the results demonstrated that the method can effectively detect and classify MPQDS.