The invention provides a method and a system for on-line detection of a metering abnormal state of a transformer. The method comprises the following steps of processing historical verification data ofa three-phase transformer of a high-voltage transmission line node to determine an effective value of a secondary side voltage or current; carrying out principal component analysis on the effective value to determine a principal component subspace, a residual subspace, the number of principal components and a Q statistic control threshold value of the secondary voltage or current of the three-phase transformer; computing Q statistic of an effective value of a secondary side voltage or current signal of the three-phase transformer, which is collected in real time; determining whether the three-phase transformer is in an abnormal state or not through comparison of the Q statistic with the control threshold; and finally computing a contribution rate of measured data of each phase of the transformer to the Q statistic, and judging an abnormal phase of the three-phase transformer according to the contribution rate. According to the method and the system, a gradual change metering abnormalstate under an online running condition of the transformer can be effectively detected, and the abnormal transformer can be accurately identified.