The invention relates to the technical field of intelligent electric meter fault diagnosis, especially to an intelligent electric meter fault diagnosis method and device based on adaptive neuro-fuzzyinference. The device includes a memory, a processor and a computer program which is stored in the memory and can run on the processor. The processor obtains at least two characteristic parameters from power consumption, electric quantity and maximum demand of an electric meter when executing a program; carrying out initial value processing of structural parameters of an adaptive neuro-fuzzy inference system (ANFIS) on characteristic parameter data composed of the continuous characteristic parameters, and determining training parameters; and training the ANFIS corresponding to the characteristics parameter data, when the training is finished, corresponding ANFIS diagnosis networks are given, and carrying out comprehensive diagnosis on the ANFIS diagnosis networks according to acquired electric meter event record data to obtain diagnosis results. The intelligent electric meter fault diagnosis method and device can automatically analyze and classify faults of the intelligent electric meters and solve the problem of insufficient analysis and classification methods of the intelligent electric meter.