The invention discloses an abnormal data diagnosis method in online monitoring of a transformer oil chromatography. The method comprises the following steps: acquiring historical data monitored by thetransformer oil chromatography and dividing the historical data according to a preset time period and gas types; pre-processing the historical data corresponding to each of the gas types, so as to obtain a pre-processing sample; obtaining a unitary frequency distribution sequence and a binary frequency distribution sequence of the pre-processing sample; then establishing an abnormal data diagnosis model; updating the unitary frequency distribution sequence and the binary frequency distribution sequence with new monitored data; finally, judging a unitary frequency and a binary frequency of to-be-monitored data according to the updated unitary frequency distribution sequence and the updated binary frequency distribution sequence and substituting the unitary frequency and the binary frequency into the abnormal data diagnosis model for diagnosis. According to the method, the diagnosis accuracy is improved favorably, and the recognition for an abnormal mode of a model is facilitated.