The hot spot temperature inside the operating power transformer directly affects its service life and operational reliability. In order to accurately evaluate the severity of the internal overheating fault of a power transformer, this paper firstly used the theory of enthalpy change to propose a novel method for calculating the fault energy coefficient, and assigned a certain weight coefficient to the fault; Secondly, based on the concept of energy coefficient, different degrees of thermal failures for various types of power transformers were the research objects, and the severity level evaluation criteria were established. This benchmark was based on the energy coefficient values calculated from the chromatographic analysis data of the 563 groups of power transformer thermal faults as the sample data, and the principle of maximum entropy was used to obtain the most consistent actual distribution of power transformer thermal fault energy coefficient probability distribution function, and the quantile values Q20, Q70 and Q95 were calculated as the graded evaluation value of the thermal fault severity. Finally, the method proposed in this paper could not only realize the quantitative assessment of the severity of thermal faults, but also reasonably divide the severity levels through analyzing the actual calculation examples, which can provide references for maintenance personnel to formulate and implement maintenance plans.