With respect to the load speciality of distribution transformers and uncertainty of thermal transmission differential equation parameters, a thermal evaluation of distribution transformers is proposed, in which the dynamic process of thermal transmission in distribution transformers is analyzed, and an idea that traditional neural network is hard to go for thermal evaluation of distribution transformers is raised. When considering three-phase load factors of distribution transformers, the LSTM Type Recurrent Neural Network model of distribution transformers thermal evaluation are established, the top layer oil temperature of distribution transformers is predicted so that the thermal evaluation of distribution transformers are achieved. The calculating example indicates that the predicting outcome method this paper used goes more closer to the practical situation than thermal transmission differential equation, in this way the validity and practicability of the method in the paper is verified.