In Japan, most of the damage caused by large earthquakes is concentrated within areas less than about 100 km from the focal regions. We have developed an earthquake alarm system that determines earthquake parameters within a few seconds of the P wave’s arrival at the closest station, and then transmits the earthquake information before the S-wave arrival in areas of possible serious earthquake damage. Since an earthquake alarm system requires the determination of reliable earthquake parameters as quickly as possible, it is unreasonable to wait until waveform data from numerous stations have been collected for analysis. For this reason, we have developed a novel method of determining the hypocentral location by using the arrival times for only a few stations, as well as the lack of P-wave arrivals at other stations at a given time. The use of not-yet-arrived data makes it possible not only to determine reliable hypocenter parameters within a few seconds, but also to detect erroneous arrival-time readings and remove them automatically. Since the available waveform data increases with time, the system was designed to redetermine earthquake parameters every second. Our method was deployed in a real-time system starting in July 2002. The real-time system locates 10–20 events per day, which includes a few felt earthquakes occurring in and around Japan. It was shown from the waveform data for about 100 felt earthquakes that almost all the events, except those far from the network, could be located within a few seconds, when most of seismic energy has not even arrived at some stations close by. We recently started widespread broadcasting of earthquake information in real-time by using a satellite transmission system.