To address problems of cloud phase retrieval using the threshold method, a cloud phase retrieval approach based on Self-Organizing Feature Map (SOFM) neural network was proposed. Cloud phase retrieval experiments were conducted using FengYun-3A/Visible and InfRared Radiometer (FY-3A/VIRR) multi-channel images, which cover the China’s territory. Experiment results indicated that the results from the SOFM neural network approach and the K-means method have good consistency, and the retrieval accuracy of the SOFM neural network exceeds that of the FY-3A operational product. Additionally, retrieval time consumed by the SOFM neural network approach is only about one third of that of the FY-3A operational product.