When the off-line transformer is re-energized, the phase-controlled switching strategy can avoid the generation of inrush current by controlling the phase. To determine the closing phase, the residual flux density (Br) in the transformer core needs to be accurately measured. This paper proposes a Br estimation method for three-phase transformers based on the finite element method and BP neural network. Firstly, the direction of Br in each phase core is determined based on the transient current characteristics. Then, the three-phase transformer is simulated and the BP neural network is trained to estimate the Br based on the simulation results. The experimental results on a three-phase transformer show that the proposed method can accurately determine the direction and amplitude of Br in each phase of the three-phase transformer.