This paper presents the final development of an expert system utilizing a measurement of cable screen earthing transient current. The developed system allows for identification and location of earth fault in underground cable and mixed lines (underground cable and overhead line) and monitoring of an earthing system and cable screen connections. The unique feature of the developed earth fault locating system is the possibility of identification of line type and branch of the MV underground cable or mixed feeder under earth fault conditions. As a result, the time to remove failure can be greatly reduced and the number of earth fault indicators installed in the distribution network can also be reduced. Unfortunately, in order to operate properly, the previously developed system requires a fundamental—50 Hz component of the measured zero-sequence cable core current and cable screen earthing current; therefore, short transient earth faults without steady-state earth fault currents cannot be localized and categorized even though the transient earth faults have a negative impact on the power system. According to measurements performed by the authors, transient earth faults are relatively frequent, which causes stress to insulation. The number of transient earth faults may be reduced by ensuring proper maintenance of the distribution system. Unfortunately, because of the very large area of the distribution feeder, often in the range of tens of kilometers or even around a hundred kilometers, and many potential causes of earth faults, it is very difficult to localize the transient earth fault and determine the cause of the earth fault. Herein, we present the possible causes of transient earth faults and methods developed for the analysis of transient earth faults. Moreover, the novel algorithm for transient earth fault detection and location is proposed. The proposed algorithm has a self-learning capability and can identify branches of the distribution feeder under transient earth fault conditions. The effectiveness of the proposed algorithm is confirmed thanks to the performed network experiment.