The invention discloses a transformer fault diagnosis method based on a krill herd optimization support vector machine, and relates to the technical field of power transformer fault diagnosis. The method comprises the following steps: normalizing collected DGA full data of a transformer and ratios of typical gases, and optimizing DGA characteristic quantity through BPSO to obtain a new DGA characteristic quantity combination; optimizing a penalty factor C and a kernel function parameter sigma of the SVM by adopting a KH optimization algorithm, constructing support vector machine diagnosis model which can be stably used, and using the model to diagnose dissolved gas in oil of the power transformer, so that comparison with a fault result obtained by expert experience is made, a final fault type is obtained, and preliminary judgment on the operation state of the transformer is realized.