Efficient and accurate fault diagnosis for power transformers can effectively protect the power system safe and stable operation. Transformer fault diagnosis method was proposed based on improved quantum-behaved particle swarm optimization fuzzy clustering to improve the accuracy of transformer fault diagnosis. Quantum-behaved particle swarm optimization algorithm was improved by genetic hybrid algorithm to improve the convergence speed and prevent falling into local extremum, which compensated the lack of sensitive of fuzzy clustering algorithm to initial value, and then the transformer achieved efficient and rapid fault diagnosis. In this algorithm, the dissolved gas in oil was taken as the characteristic quantity of fault, the fault set was composed of 68 groups of data, and improved QPSO was adopted to obtain the optimal initial cluster centers for verifying the 3 different data sets. Experiments show that the effectiveness of the proposed method.