The purpose of most earthquake early warning systems (EEWSs) is promptly to identify damaging earthquakes occurring at local distances from a seismic station or network. However, mislocation of a large regional or teleseismic event that is relatively innocuous to the local society can cause an EEWS to issue a serious false alarm. To solve this problem, we applied the recently developed, robust maximum‐likelihood earthquake location method (MAXEL) to the EEWS in South Korea, which is adjacent to several major plate boundaries capable of producing very large earthquakes at regional distances. MAXEL, which is based on the maximum‐likelihood method using the concepts of equal differential times of P arrivals and Student’s t distribution, was optimized for the application and event identification criteria were introduced. Although the maximum number of arrivals for locating an earthquake was controlled to allow for fast computation and successful operation of the EEWS, it was possible to maintain location accuracy. Synthetic tests based on current Korean seismic networks showed that MAXEL could locate local events very well and also handle regional earthquakes effectively. In addition, offline tests using observed data also clearly confirmed the robustness of MAXEL to outliers and its superior capability for discriminating local events from regional or teleseismic events. Therefore, it is expected that MAXEL could significantly improve the stability and accuracy of the phase association and event location of EEWSs and that MAXEL has great potential for earthquake early warning.