Association rule mining methods are commonly utilized to analyze the dissolved gas which is applied to diagnose the power transmission fault events. For the purpose of improving the performance, this paper proposes a diagnosis method for power transformer fault events based on the enhanced association rule mining algorithm. Firstly, the conditional significance measurements which can be adapted for different input features are established. Thus the rarely distributed but risky data can be incorporated in analysis, and all the potential circumstances in reality can be considered. Next, the risk weights of input data are generated through their likelihood of causing a fault rather than their statistical distribution. Therefore, the impact of each input will be measured more precisely. Finally, Relim algorithm is applied to raise the efficiency of mining. The empirical study shows that the proposed method is more pinpoint, realizable and efficient compared with the methods with the fixed significance measurements, the conventional technique to calculate the risk weight, and Apriori algorithm.