The invention relates to the power device detection technology field, especially relates to a testing method of an analog-digital circuit fault diagnosis based on a neural network expert system. The neural network can be combined with the expert system by adopting the neural network to build the expert system, and the inference of the conventional expert system based on the symbols can be turned to the inference based on the numeric calculation, which can be used for the analog-digital mixed circuit fault diagnosis. The testing method comprises the following steps such as establishing of ANNES knowledge base, establishing and training of the BP network, inference and diagnosis. The testing method is advantageous in that the execution efficiency of the expert system can be increased, the intelligent level of the system can be raised, the system performance can be improved, and the conclusion can be directly achieved by the output of the neural network; the testing process can be greatly simplified, the testing time and the testing costs can be saved, the working efficiency can be obviously increased, and the testing method is suitable for the actual engineering application.