Earthquake early warning (EEW) systems systematically monitor ground motion and provide alerts prior to damaging shaking. However, these systems present certain challenges, which include avoiding alerts caused by false triggers or delayed alerts for earthquakes from outside the seismic network. Here, we show that an array processing approach, which is capable of identifying the propagation direction and type of seismic phase, can overcome these challenges. In this study, a new automatic array processing approach was implemented using a single mini array to determine the back azimuth and slowness of the phases for local events. Seismograms of local events with ML 2.0–4.6 that occurred in the Korean Peninsula were analyzed using three array processing methods: beam packing, frequency–wavenumber, and plane-wave fitting. In addition, a combined P-wave picking procedure was introduced, and its performance was evaluated. We developed the criteria associated with slowness range and consistency of back azimuths to determine a stable back azimuth from the three array processing methods, estimating the back azimuth of the seismic phase with more accuracy than any single method. It is expected that seismic mini arrays and the developed automatic array processing approach can be used for network EEW systems to discriminate alarms caused by false triggers. In addition, the developed approach can be used for rapid onsite and front-detection warnings using the array inside or on the outskirts of a network when combined with other estimations, such as an epicentral distance or another back azimuth.