The invention provides a KNN-based distribution network ground fault identification method comprising steps of obtaining the zero-sequence current waveform data of different fault causes through an artificial grounding test or field actual fault recording, extracting five feature parameters representing the different fault causes and including the self-recovery, the transition time, the zero-current time, the distortion degree and the random degree to form a feature sample database; and when a fault occurs, calculating the above features of an input waveform and searching for a cause categoryof the fault by an KNN algorithm. The method, on the one hand, can focus on the fault cause before fault inspection so as to formulate a targeted inspection scheme, greatly improve fault finding efficiency and reduce power outage loss; on the other hand, can accurately obtain the distribution law of the distribution network fault causes of a certain supply area, creates hidden danger treatment andfault prevention measures, promotes the improvement in the level of distribution network operation and maintenance management, and improves the feature sample database with an increase in the numberof fault samples and features so as to achieve a high recognition success rate.