An image derived by stacking long-period seismograms is used as an empirical matched filter to detect and locate earthquakes. Records from 564 events (mb≥6) recorded at very long periods (T≥60 s) by the 15 to 20 stations of the International Deployment of Accelerometers (IDA) network are stacked using a method that emphasizes the surface wave arrivals. The first 3 hours of this time versus range image are used to construct a matched filter for continuous application of the IDA data. The output of this filter contains spatial and temporal peaks that define the location and origin time of probable seismic events. Implementation of this technique to 11 years of IDA data from 1981 to 1991 identifies 4061 events. These include 65% of cataloged events of mb≥5.5. Earthquakes which appear anomalously strong relative to their surface wave magnitudes are mainly located on oceanic transform faults and probably represent unusually slow ruptures. The method successfully detects two eruptions of the El Chich¿n volcano in southern Mexico, from the anomalous low-frequency energy radiated during the eruptions. In addition, 32 earthquakes are detected which are not in the standard global catalogs. These events appear to be about Ms=5 and are mainly located in the southern oceans, where there are gaps in the coverage of the high-frequency networks. Although these earthquakes are probably ''slow'' since they occur mostly on oceanic transform faults, they are not ''silent'' as they also can be observed in higher-frequency data. ¿ American Geophysical Union 1994