A seismic antenna approach based on the generalized zero-lag cross-correlation method for rapid earthquake localization is proposed. This method is intended to be applied primarily for early warning, whenever the epicentre-to-target distances guarantee enough lead-time, rapid response purposes, and for those circumstances when a seismogenic area is not directly accessible with seismic stations or/and a network of instruments is concentrated within the area to be warned. The procedure we propose aims to provide useful information for magnitude determination and shake-maps generation. Indeed, it relies only on the first P-wave triggered arrivals from seismic stations, and is designed to work in real-time for the localization of events occurring outside of the network, that is, under conditions that might be detrimental to standard localization approaches. The procedure can by summarized by a few preliminary pre-seismic and real-time co-seismic steps. In the pre-seismic time-frame, for the cases where a large and dense network exists, waiting for all stations to trigger could dramatically reduce the available lead-time for the warning. Therefore, in such cases, the network could profitably be divided into sub-arrays, while also taking advantage of available earthquake recordings or simulated data sets. During the co-seismic time-frame, the main operations are: (1) individual on-site triggering by the P-wave of the seismic stations (e.g. by a STA/LTA algorithm); (2) real-time communication of key parameters (e.g. P-wave arrival time, and signal quality) to a main centre by SMS/WLAN; (3) setup of a pseudo data set, composed by a Gaussian function centred at the P-time, and with a bell width that can be set up proportional to the trigger signal-to-noise ratio (SNR); (4) calculation of a coherency map for the sub-array with triggered stations (preliminary sub-array location); and (5) stacking of coherency maps from the different sub-arrays (final location). By the stack of coherency maps estimated by the different sub-arrays in the last step of the procedure, the epicentral area’s location may be better constrained. This innovative approach for rapid localization was applied to both synthetic data, and real observations of two small earthquakes that occurred in the Marmara Sea, Turkey, which were recorded by the Istanbul Earthquake Rapid Response System. ©2011 Springer Science+Business Media B.V.