The existing early warning methods for the temperature field of roller compacted concrete (RCC) dams are mostly based on a single measuring point which is lack of the focus on the whole temperature field. To solve this problem, this paper developed a new method using the comprehensive information of multiple measuring points. Firstly, the order degree and disorder degree of the monitored temperature were defined by probability method. Then the weight distribution of each measuring point was determined using the projection pursuit method. Secondly, according to the evolution law of the temperature field, the weight entropy, which can describe the order degree of the temperature field, was constructed based on the order degree and weight distribution. Thirdly, the early warning criteria were proposed using the typical small probability principle. The case analysis shows that the change law of weight entropy is consistent with that of the measured temperature, so the feasibility and effectiveness of the method are verified.