We develop an algorithm for the detection and location of seismic sources using intermediate-period (35–150 sec) surface waves recorded on a global array of stations. Continuous vertical seismic waveforms from the global network are collected and a 4° × 4° global grid of target locations is defined. For each target location and each station, a surface-wave propagation operator is deconvolved from the seismogram to restore any source pulse present. The envelope of the seismogram is calculated and cross correlated with a theoretical source-pulse shape. The resulting waveforms are stacked to improve signal-to-noise characteristics, and the quality, strength, and timing of the potential detection are determined. When a successful event detection is made, a finer grid is applied to locate the event with greater precision. We apply the algorithm systematically for the period 1993–2003 and catalog the events. Approximately 2000 events are detected and located for each year and 98% of shallow events in the Harvard Centroid Moment Tensor (cmt) catalog are detected and located by the new algorithm. A comparison of 9482 events common to the two catalogs allows the detection strength to be calibrated against the cmt moment magnitude. All detected events have estimated moment magnitudes Mw >4.5. In each year, approximately 100 events not listed in other global seismicity catalogs are detected and located. Many of these events lie along the ridge-transform plate boundaries in the Southern Hemisphere and may be regular earthquakes that have gone undetected because of poor station coverage. Other events, located in areas where global and regional networks provide good coverage, are potentially anomalous and may have escaped detection as a result of their unusual source properties.