Current differential relay is commonly used to protect power transformer. However, current differential relay will be tripod by judging like internal fault during inrush occurring in transformer. To resolve such problem, this paper proposes a new protective relaying algorithm using Neuro-Fuzzy Inference. A variety of transformer transition states are simulated by BCTRAN and HYSDT of EMTP. Primary phase voltage and differential current are obtained from simulation. The target data which are used in Neuro-Fuzzy algorithm are obtained from transformed primary voltage and current. Then, these are trained by Neuro-Fuzzy algorithm. The trained Neuro-Fuzzy algorithm correctly distinguishes whether internal fault occurs or not, within 1/2 cycle after fault. Accordingly, it is evaluated that the proposed algorithm has good relaying characteristics.