The invention provides a transformer fault diagnosis method on the basis of weighted gray correlation and fuzzy clustering. The transformer fault diagnosis method includes steps of acquiring dissolved gas concentration value in a transformer in real time, wherein dissolved gases include acetylene, ethylene, methane, hydrogen and ethane; determining a cluster of a data matrix by means of a fuzzy cluster algorithm, and determining a fault sequence according to a cluster result; analyzing the weighted gray correlation of a sequence to be detected and the fault sequence to finally obtain a sequence with the highest correlation to the sequence to be detected, so that fault types corresponding to the sequence are those which the sequence to be detected belongs to, and a transformer fault diagnosis result is obtained. The transformer diagnosis method can obtain accurate fault diagnosis under the condition of low confidence level, reference is provided for reasonable maintenance and safe operation, waste is reduced, correlation is weighted to be calculated through the gray correlation analysis method, and influences on faults due to different factors can be differentiated, so that workload can be greatly reduced during fault removal, and labor resources can be greatly saved.