This paper described a new method of fault location detection for 20 kV of electrical distribution system using an injection of voltage impulse. The impulse device injects pulses into the network every second in each phase and at the same time records the reflection signals. When a disturbance occurs in the network, there will be a new discontinuity or impedance mismatch which causes the reflection signal to change. A comparative method is used to compare signals before and after the fault. From, the comparison result we will get the time difference (Δt) which is calculated at the first signal injected until both of them separated each other. The fault distance (d) obtained by multiplying the signal speed in the medium by half of the time difference that resulted from signals comparison. Furthermore, the result of the time difference and fault distance will be learned using Adaptive Neuro-Fuzzy Inference System (ANFIS) to show an improved method of fault location detection. Simulation is done using MATLAB with some conditions, such as: varying the fault type (one phase to ground (LG), two phases to the ground (LLG), and three phases to the ground (LLLG)), fault resistance (Rf), ground resistance (Rg), and fault location. Simulation results using ANFIS show an improvement in the accuracy of fault location detection, for example, 1,853.35 m of distance, when a 3-phase fault applied, the distance error drops from 59.13 m (3.19%) to 13.49 m (0.73%).