In power grid

many on-line monitoring devices of transformer bushings have been used. However

moni-toring data not only depend on insulation condition of the bushing

but also are affected by environmental condition

including such as temperature

ambient humidity and working voltage. In addition

the existing early warning strate-gies recommended by different bushing manufactures are diverse

which brings difficulty to the application of on-line monitoring. Therefore

proper warning strategy is essential for diagnosis of insulation defects. In this paper

themathematical statistics are used to analyze the historical on-line monitoring data of in-service transformer bushings inGuangdong Power Grid

and about 1% of data exception are found

which cause false alarm. After removing these ab-normal data

the data sample fluctuates slightly around the mean value within the range of ±15% and is subject to ap-proximate normal distribution. Furthermore

a novel warning strategy

i.e.

the relative offset of the dissipation factoris limited within the range of 30%

is proposed and tested. The new warning strategy adapts to different rated dielec-tric loss of various types of transformers

and has better applicability.