Abstract: Temperature stands as a pivotal factor influencing the insulation of transformer oil. To realize non-destructive testing of the internal temperature distribution of the transformer, a new ultrasonic temperature detection method is proposed based on unidirectional and reflective joint measurement, which enables accurate determination of the ultrasonic flight time along the propagation paths inside the transformer. Additionally, a three-dimensional temperature field visualization algorithm for transformers is proposed, relying on improved Markov radial basis function (Markov-RBF). The transformer’s interior is partitioned into several sub-blocks, and an algorithm is utilized to fit the overall temperature by averaging the temperatures at the center points of these sub-blocks. This algorithm effectively addresses the challenge of inaccessible temperature data in areas where ultrasonic wave propagation is hindered within the transformer. It also resolves the ill-conditioned matrix issue caused by the number of propagation paths being fewer than the number of sub-blocks. Consequently, the reconstruction of the three-dimensional temperature field is successfully accomplished inside the transformer, enabling the visualization of temperature distributions and localization of the highest temperature. Taking a 10 kV three-phase three-column oil-immersed experimental transformer as an example, COMSOL simulation is utilized to compare and analyze the temperature field reconstruction results of the detection system. The results demonstrate a consistency between the two reconstructed temperature fields, with the highest oil temperature occurring near the winding at 70% of the tank height, thereby validating the effectiveness of this method.