When a single-phase-to-ground fault occurs in the small current system, its fault characteristics are easily affected by weak fault conditions such as the high grounding transition resistance and the small initial phase angle. Therefore, this paper presents a method of the single-phase-to-ground fault line selection based on an improved Hilbert?Huang transform-random forest. Firstly, the current transient signals of every lines are extracted. Then the pure transient electrical quantities are extracted by the improved Hilbert?Huang transform, and three kinds of eigenvectors such as standard deviations, energy entropy and amplitude distortion degrees are constructed. In the following, the eigenvectors are input into the random forest classifier to establish a fault line selection model, and the fault line selection problem is then transformed into a binary classification problem which realizing the automatic identification of fault lines. The simulation results show that the proposed method can effectively improve the accuracy of fault line selection by comprehensively using the amplitude, frequency and energy of transient signal; whatsmore it is not affected by weak fault condition and feeder structure, it hence has strong adaptability and reliability.